diff --git a/tests/common_faiss_tests.py b/tests/common_faiss_tests.py
index 8dc25edec0..a8afe344e4 100644
--- a/tests/common_faiss_tests.py
+++ b/tests/common_faiss_tests.py
@@ -49,7 +49,6 @@ def evalres(self, DI):
         for rank in 1, 10, 100:
             e[rank] = ((I[:, :rank] == self.gt.reshape(-1, 1)).sum() /
                        float(self.nq))
-        # print("1-recalls: %s" % e)
         return e
 
 
diff --git a/tests/test_RCQ_cropping.cpp b/tests/test_RCQ_cropping.cpp
index 4dd3470885..4463c256ed 100644
--- a/tests/test_RCQ_cropping.cpp
+++ b/tests/test_RCQ_cropping.cpp
@@ -28,7 +28,6 @@ TEST(RCQCropping, test_cropping) {
     faiss::ResidualCoarseQuantizer rcq(d, nbits);
 
     rcq.train(nt, xt);
-    // fprintf(stderr, "nb centroids: %zd\n", rcq.ntotal);
 
     // the test below works only for beam size == nprobe
     rcq.set_beam_factor(1.0);
@@ -44,7 +43,6 @@ TEST(RCQCropping, test_cropping) {
     nbits.pop_back();
     faiss::ResidualCoarseQuantizer rcq_cropped(d, nbits);
     rcq_cropped.initialize_from(rcq);
-    // fprintf(stderr, "cropped nb centroids: %zd\n", rcq_cropped.ntotal);
 
     EXPECT_EQ(rcq_cropped.ntotal, rcq.ntotal >> last_nbits);
 
diff --git a/tests/test_clustering.py b/tests/test_clustering.py
index 2b81fc3e35..b1afc8523f 100644
--- a/tests/test_clustering.py
+++ b/tests/test_clustering.py
@@ -110,9 +110,6 @@ def test_weighted(self):
         cdis2_first = cdis2[:5].sum()
         cdis2_last = cdis2[5:].sum()
 
-        print(cdis1_first, cdis1_last)
-        print(cdis2_first, cdis2_last)
-
         # with the new clustering, the last should be much (*2) closer
         # to their centroids
         self.assertGreater(cdis1_last, cdis1_first * 2)
diff --git a/tests/test_contrib.py b/tests/test_contrib.py
index 0e7cbbfb03..05a2c4ac8b 100644
--- a/tests/test_contrib.py
+++ b/tests/test_contrib.py
@@ -147,7 +147,6 @@ def test_query_iterator(self, metric=faiss.METRIC_L2):
         xb = ds.get_database()
         D, I = faiss.knn(xq, xb, 10, metric=metric)
         threshold = float(D[:, -1].mean())
-        print(threshold)
 
         index = faiss.IndexFlat(32, metric)
         index.add(xb)
@@ -251,7 +250,6 @@ def test_precision_recall(self):
         Inew = np.hstack(Inew)
 
         precision, recall = evaluation.range_PR(lims_ref, Iref, lims_new, Inew)
-        print(precision, recall)
 
         self.assertEqual(precision, 0.6)
         self.assertEqual(recall, 0.6)
diff --git a/tests/test_contrib_with_scipy.py b/tests/test_contrib_with_scipy.py
index cb81bb623c..4f89e2fc1b 100644
--- a/tests/test_contrib_with_scipy.py
+++ b/tests/test_contrib_with_scipy.py
@@ -44,7 +44,6 @@ def test_sparse_routines(self):
         faiss.normalize_L2(xt)
 
         mask = np.abs(xt) > 0.045
-        # print("fraction:", mask.sum() / mask.size) # around 10% non-zeros
         xt[np.logical_not(mask)] = 0
 
         centroids = ds.get_queries()
@@ -72,7 +71,6 @@ def test_sparse_kmeans(self):
         faiss.normalize_L2(xt)
 
         mask = np.abs(xt) > 0.045
-        # print("fraction:", mask.sum() / mask.size) # around 10% non-zeros
         xt[np.logical_not(mask)] = 0
 
         km = faiss.Kmeans(ds.d, 50)
diff --git a/tests/test_fast_scan.py b/tests/test_fast_scan.py
index b061ee3af0..cfe9636fee 100644
--- a/tests/test_fast_scan.py
+++ b/tests/test_fast_scan.py
@@ -34,7 +34,6 @@ def test_PQ4_accuracy(self):
         nq = Iref.shape[0]
         recall_at_1 = (Iref[:, 0] == Ia[:, 0]).sum() / nq
         assert recall_at_1 > 0.6
-        # print(f'recall@1 = {recall_at_1:.3f}')
 
 
     # This is an experiment to see if we can catch performance
@@ -498,7 +497,6 @@ def subtest_accuracy(self, aq, st, implem, metric_type='L2'):
         recall_ref = (Iref == gt).sum() / nq
         recall = (Ia == gt).sum() / nq
 
-        print(aq, st, implem, metric_type, recall_ref, recall)
         assert abs(recall_ref - recall) < 0.05
 
     def xx_test_accuracy(self):
@@ -531,7 +529,6 @@ def subtest_from_idxaq(self, implem, metric):
         nq = Iref.shape[0]
         recall_ref = (Iref == gt).sum() / nq
         recall1 = (I1 == gt).sum() / nq
-        print(recall_ref, recall1)
         assert abs(recall_ref - recall1) < 0.05
 
     def xx_test_from_idxaq(self):
diff --git a/tests/test_index_composite.py b/tests/test_index_composite.py
index a760c0cf09..8d9b441adc 100644
--- a/tests/test_index_composite.py
+++ b/tests/test_index_composite.py
@@ -168,8 +168,6 @@ def test_remove_id_map_2(self):
         index.remove_ids(remove_set)
         index.add_with_ids(X[5:, :], idx[5:])
 
-        print (index.search(X, 1))
-
         for i in range(10):
             _, searchres = index.search(X[i:i + 1, :], 1)
             if idx[i] in remove_set:
@@ -954,7 +952,6 @@ def do_test(self, factory_string):
         index.nprobe = 10
         Dref, Iref = index.search(ds.get_queries(), 10)
 
-        #print(index.search_and_return_codes)
         D, I, codes = index.search_and_return_codes(
             ds.get_queries(), 10, include_listnos=True)
 
diff --git a/tests/test_local_search_quantizer.py b/tests/test_local_search_quantizer.py
index 01fec70ccf..7975929811 100644
--- a/tests/test_local_search_quantizer.py
+++ b/tests/test_local_search_quantizer.py
@@ -196,7 +196,6 @@ def test_update_codebooks_with_double(self):
         err_float = eval_codec(lsq, xb)
 
         # 6533.377 vs 25457.99
-        print(err_double, err_float)
         self.assertLess(err_double, err_float)
 
     def test_compute_binary_terms(self):
@@ -348,7 +347,6 @@ def test_training(self):
         pq.train(xt)
         err_pq = eval_codec(pq, xb)
 
-        print(err_lsq, err_pq)
         self.assertLess(err_lsq, err_pq)
 
 
@@ -463,7 +461,6 @@ def eval_index_accuracy(self, factory_key):
             index.nprobe = nprobe
             D, I = index.search(ds.get_queries(), 10)
             inter = faiss.eval_intersection(I, ds.get_groundtruth(10))
-            # print("nprobe=", nprobe, "inter=", inter)
             inters.append(inter)
 
         inters = np.array(inters)
@@ -528,7 +525,6 @@ def test_codec(self):
         pq.train(xt)
         err_pq = eval_codec(pq, xb)
 
-        print(err_plsq, err_pq)
         self.assertLess(err_plsq, err_pq)
 
     def test_with_lsq(self):
@@ -549,7 +545,6 @@ def test_with_lsq(self):
         lsq.train(xt)
         err_lsq = eval_codec(lsq, xb)
 
-        print(err_plsq, err_lsq)
         self.assertEqual(err_plsq, err_lsq)
 
     def test_lut(self):
@@ -664,7 +659,6 @@ def test_index_accuracy2(self):
         """check that the error is in the same ballpark as LSQ."""
         inter1 = self.eval_index_accuracy("IVF32,PLSQ2x2x5_Nqint8")
         inter2 = self.eval_index_accuracy("IVF32,LSQ4x5_Nqint8")
-        # print(inter1, inter2)  # 381 vs 374
         self.assertGreaterEqual(inter1 * 1.1, inter2)
 
     def test_factory(self):
diff --git a/tests/test_partition.py b/tests/test_partition.py
index 02de7e8c2c..fd41eabe1f 100644
--- a/tests/test_partition.py
+++ b/tests/test_partition.py
@@ -49,7 +49,6 @@ def do_partition(self, n, q, maxval=None, seed=None):
         if seed is None:
             for i in range(50):
                 self.do_partition(n, q, maxval, i + 1234)
-        # print("seed=", seed)
         rs = np.random.RandomState(seed)
         if maxval is None:
             vals = rs.rand(n).astype('float32')
@@ -95,7 +94,6 @@ def do_partition(self, n, q, maxval=None, seed=None):
         if seed is None:
             for i in range(50):
                 self.do_partition(n, q, maxval, i + 1234)
-        # print("seed=", seed)
         rs = np.random.RandomState(seed)
         if maxval is None:
             vals = rs.rand(n).astype('float32')
@@ -148,7 +146,6 @@ def do_partition(self, n, q, maxval=65536, seed=None):
             for i in range(50):
                 self.do_partition(n, q, maxval, i + 1234)
 
-        # print("seed=", seed)
         rs = np.random.RandomState(seed)
         vals = rs.randint(maxval, size=n).astype('uint16')
         ids = (rs.permutation(n) + 12345).astype('int64')
@@ -160,7 +157,6 @@ def do_partition(self, n, q, maxval=65536, seed=None):
         tab_a = faiss.AlignedTableUint16()
         faiss.copy_array_to_AlignedTable(vals, tab_a)
 
-        # print("tab a type", tab_a.get())
         if type(q) == int:
             faiss.CMax_uint16_partition_fuzzy(
                 tab_a.get(), sp(ids), n, q, q, None)
@@ -196,7 +192,6 @@ def do_partition(self, n, q, maxval=65536, seed=None):
         if seed is None:
             for i in range(50):
                 self.do_partition(n, q, maxval, i + 1234)
-        # print("seed=", seed)
         rs = np.random.RandomState(seed)
         vals = rs.randint(maxval, size=n).astype('uint16')
         ids = (rs.permutation(n) + 12345).astype('int64')
@@ -209,7 +204,6 @@ def do_partition(self, n, q, maxval=65536, seed=None):
         vals_inv = (65535 - vals).astype('uint16')
         faiss.copy_array_to_AlignedTable(vals_inv, tab_a)
 
-        # print("tab a type", tab_a.get())
         if type(q) == int:
             faiss.CMin_uint16_partition_fuzzy(
                 tab_a.get(), sp(ids), n, q, q, None)
diff --git a/tests/test_product_quantizer.py b/tests/test_product_quantizer.py
index 1cdee7f144..e05426b129 100644
--- a/tests/test_product_quantizer.py
+++ b/tests/test_product_quantizer.py
@@ -26,7 +26,6 @@ def test_pq(self):
         x2 = pq.decode(codes)
         diff = ((x - x2)**2).sum()
 
-        # print("diff=", diff)
         # diff= 4418.0562
         self.assertGreater(5000, diff)
 
diff --git a/tests/test_residual_quantizer.py b/tests/test_residual_quantizer.py
index e37ee3efe2..f87e7650d9 100644
--- a/tests/test_residual_quantizer.py
+++ b/tests/test_residual_quantizer.py
@@ -211,7 +211,6 @@ def test_training(self):
 
         # in practice RQ is often better than PQ but it does not the case here, so just check
         # that we are within some factor.
-        # print(err_pq, err_rq)
         self.assertLess(err_rq, err_pq * 1.2)
 
     def test_beam_size(self):
@@ -321,10 +320,8 @@ def retrain_AQ_codebook(index, xt):
 
     x_decoded = index.sa_decode(codes_packed)
     MSE = ((xt - x_decoded) ** 2).sum() / n
-    # print(f"Initial MSE on training set: {MSE:g}")
 
     codes = unpack_codes(index.rq, codes_packed)
-    # print("ref codes", codes[0])
     codebook_offsets = faiss.vector_to_array(rq.codebook_offsets)
 
     # build sparse code matrix (represented as a dense matrix)
@@ -343,7 +340,6 @@ def retrain_AQ_codebook(index, xt):
         B, residuals, rank, singvals = scipy.linalg.lstsq(C, xt, )
 
     MSE = ((C @ B - xt) ** 2).sum() / n
-    # print(f"MSE after retrainining: {MSE:g}")
 
     # replace codebook
     # faiss.copy_array_to_vector(B.astype('float32').ravel(), index.rq.codebooks)
@@ -503,7 +499,6 @@ def test_reestimate_codebook_2(self):
         xt_decoded = ir.sa_decode(ir.sa_encode(xt))
         err_after_refined = ((xt - xt_decoded) ** 2).sum()
 
-        # print(err_before, err_after_refined)
         # ref run 7474.98 / 7006.1777
         self.assertGreater(err_before, err_after_refined * 1.06)
 
@@ -781,7 +776,6 @@ def test_search_L2(self):
             else:
                 inter_2 = faiss.eval_intersection(I2, gt)
                 self.assertGreaterEqual(inter_ref, inter_2)
-                # print(st, inter_ref, inter_2)
 
 
 ###########################################################
@@ -814,7 +808,6 @@ def do_test_accuracy(self, by_residual, st):
             index.nprobe = nprobe
             D, I = index.search(ds.get_queries(), 10)
             inter = faiss.eval_intersection(I, ds.get_groundtruth(10))
-            # print(st, "nprobe=", nprobe, "inter=", inter)
             inters.append(inter)
 
         # do a little I/O test
@@ -909,18 +902,13 @@ def do_test_accuracy_IP(self, by_residual):
             D, I = index.search(ds.get_queries(), 10)
             index.rq.search_type = faiss.AdditiveQuantizer.ST_LUT_nonorm
             D2, I2 = index.search(ds.get_queries(), 10)
-            # print(D[:5] - D2[:5])
-            # print(I[:5])
             np.testing.assert_array_almost_equal(D, D2, decimal=5)
             # there are many ties because the codes are so short
             self.assertLess((I != I2).sum(), I.size * 0.1)
 
             # D2, I2 = index2.search(ds.get_queries(), 10)
-            # print(D[:5])
-            # print(D2[:5])
 
             inter = faiss.eval_intersection(I, ds.get_groundtruth(10))
-            # print("nprobe=", nprobe, "inter=", inter)
             inters.append(inter)
         self.assertTrue(np.all(inters[1:4] >= inters[:3]))
 
@@ -1166,7 +1154,6 @@ def test_codec(self):
         pq.train(xt)
         err_pq = eval_codec(pq, xb)
 
-        # print(err_prq, err_pq)
         self.assertLess(err_prq, err_pq)
 
     def test_with_rq(self):
@@ -1271,7 +1258,6 @@ def test_index_accuracy2(self):
         """check that the error is in the same ballpark as RQ."""
         inter1 = self.eval_index_accuracy("IVF100,PRQ2x2x5_Nqint8")
         inter2 = self.eval_index_accuracy("IVF100,RQ4x5_Nqint8")
-        # print(inter1, inter2)  # 392 vs 374
         self.assertGreaterEqual(inter1 * 1.1, inter2)
 
     def test_factory(self):
diff --git a/tests/test_rowwise_minmax.py b/tests/test_rowwise_minmax.py
index dbd14de388..53e6c00b15 100644
--- a/tests/test_rowwise_minmax.py
+++ b/tests/test_rowwise_minmax.py
@@ -45,7 +45,6 @@ def compare_train_vs_train_inplace(self, factory_key):
 
         # make sure that the reconstruction error is not crazy
         reconstruction_err = ((x - decoded) ** 2).sum()
-        print(reconstruction_err)
 
         self.assertLess(reconstruction_err, 0.6)
 
diff --git a/tests/test_standalone_codec.py b/tests/test_standalone_codec.py
index 7fdcf6849f..2176a12e99 100644
--- a/tests/test_standalone_codec.py
+++ b/tests/test_standalone_codec.py
@@ -151,7 +151,6 @@ def compare_accuracy(self, lowac, highac, max_errs=(1e10, 1e10)):
             err = ((x - x2) ** 2).sum()
             errs.append(err)
 
-        print(errs)
         self.assertGreater(errs[0], errs[1])
 
         self.assertGreater(max_errs[0], errs[0])
@@ -214,7 +213,6 @@ def test_repeats(self):
             code = repeats.encode(swig_ptr(vec))
             vec2 = np.zeros(dim, dtype='float32')
             repeats.decode(code, swig_ptr(vec2))
-            # print(vec2)
             assert np.all(vec == vec2)
 
     def test_ZnSphereCodec_encode_centroid(self):
@@ -222,7 +220,6 @@ def test_ZnSphereCodec_encode_centroid(self):
         r2 = 5
         ref_codec = faiss.ZnSphereCodec(dim, r2)
         codec = faiss.ZnSphereCodecRec(dim, r2)
-        # print(ref_codec.nv, codec.nv)
         assert ref_codec.nv == codec.nv
         s = set()
         for i in range(ref_codec.nv):
@@ -237,7 +234,6 @@ def test_ZnSphereCodecRec(self):
         dim = 16
         r2 = 6
         codec = faiss.ZnSphereCodecRec(dim, r2)
-        # print("nv=", codec.nv)
         for i in range(codec.nv):
             c = np.zeros(dim, dtype='float32')
             codec.decode(i, swig_ptr(c))
@@ -300,15 +296,10 @@ def test_rw(self):
         for i in range(nbyte):
             self.assertTrue(((bignum >> (i * 8)) & 255) == bs[i])
 
-        #for i in range(nbyte):
-        #    print(bin(bs[i] + 256)[3:], end=' ')
-        # print()
-
         br = faiss.BitstringReader(swig_ptr(bs), nbyte)
 
         for nbit, xref in ctrl:
             xnew = br.read(nbit)
-            # print('nbit %d xref %x xnew %x' % (nbit, xref, xnew))
             self.assertTrue(xnew == xref)
 
     def test_arrays(self):