diff --git a/src/core/reference/include/ngraph/runtime/reference/fake_quantize.hpp b/src/core/reference/include/ngraph/runtime/reference/fake_quantize.hpp
index 94f613cd91fb75..7982dc841d9184 100644
--- a/src/core/reference/include/ngraph/runtime/reference/fake_quantize.hpp
+++ b/src/core/reference/include/ngraph/runtime/reference/fake_quantize.hpp
@@ -19,11 +19,62 @@ namespace ngraph {
 namespace runtime {
 namespace reference {
 namespace fake_quantize_details {
+inline std::vector<size_t> calc_broadcast_index_offset(const std::vector<size_t>& memory_offsets,
+                                                       const std::vector<size_t>& broadcast_shape) {
+    std::vector<size_t> broadcast_offsets(broadcast_shape.size(), 0);
+    for (int i = static_cast<int>(broadcast_shape.size()) - 2; i >= 0; --i) {
+        if (broadcast_shape[i] == 1) {
+            broadcast_offsets[i] = memory_offsets[i];
+        }
+    }
+    const auto not_one = [](size_t i) {
+        return i != 1;
+    };
+    if (std::any_of(broadcast_shape.begin(), broadcast_shape.end(), not_one) && broadcast_shape.back() == 1) {
+        broadcast_offsets[broadcast_offsets.size() - 1] = 1;
+    }
+    if (broadcast_shape.back() == 1) {
+        for (int i = static_cast<int>(broadcast_shape.size()) - 1; i >= 0; --i) {
+            if (broadcast_shape[i] != 1) {
+                broadcast_offsets[i] = memory_offsets[i] - 1;
+                break;
+            }
+        }
+    }
+    return broadcast_offsets;
+}
 
 inline size_t calc_full_broadcast_offset(const std::vector<size_t>& current_dims, const std::vector<size_t>& offsets) {
     return std::inner_product(begin(current_dims), end(current_dims), begin(offsets), uint64_t(0));
 }
 
+inline Shape align_shape_sizes(const Shape& shape, const Shape& target_shape, const op::AutoBroadcastSpec& broadcast) {
+    Shape s;
+    switch (broadcast.m_type) {
+    case op::AutoBroadcastType::NONE: {
+        s = shape;
+        break;
+    }
+    case op::AutoBroadcastType::NUMPY: {
+        s = Shape(target_shape.size(), 1);
+        std::copy(begin(shape), end(shape), prev(end(s), shape.size()));
+        break;
+    }
+    case op::AutoBroadcastType::PDPD: {
+        const size_t axis =
+            broadcast.m_axis == -1 ? target_shape.size() - shape.size() : static_cast<size_t>(broadcast.m_axis);
+
+        s = Shape(target_shape.size(), 1);
+        const auto axis_to_copy = target_shape.size() - axis;
+        const auto b = begin(shape);
+        const auto e = b + axis_to_copy;  // from e to end(shape) should be only ones
+        std::copy(b, e, next(begin(s), axis));
+        break;
+    }
+    }
+    return s;
+}
+
 inline void increment_current_dim(std::vector<size_t>& current_dims, const std::vector<size_t>& shape) {
     size_t incremented_dim_number = current_dims.size();
     while (incremented_dim_number-- > 0) {
@@ -54,7 +105,9 @@ class QuantizationBound {
             bound = Bound::aligned;
         } else {
             bound = Bound::broadcast;
-            row_strides = row_major_strides(arg_shape);
+            const auto arg_memory_offsets = row_major_strides(arg_shape);
+            const auto unsqueezed_bound_shape = align_shape_sizes(bound_shape, arg_shape, broadcast_spec);
+            row_strides = calc_broadcast_index_offset(arg_memory_offsets, unsqueezed_bound_shape);
         }
     }
     T get_value(const std::vector<size_t>& current_dim, size_t idx) const {
diff --git a/src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp b/src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp
index 5e58623f5a6dc3..224abb78c87322 100644
--- a/src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp
+++ b/src/plugins/intel_cpu/tests/functional/shared_tests_instances/skip_tests_config.cpp
@@ -186,6 +186,9 @@ std::vector<std::string> disabledTestPatterns() {
         R"(smoke_dynamic_BatchSizeOne/RNNSequenceCPUTest.*IS=\(\[1\.\?\.10\]_\[1\.1\.10\]_\[\?\]_\)_TS=\{\(1\.2\.10\)_\(1\.1\.10\)_\(1\)\}_\{\(1\.4\.10\)_\(1\.1\.10\)_\(1\)\}_\{\(1\.8\.10\)_\(1\.1\.10\)_\(1\)\}_seqMode=PURE_SEQ_activations=\(relu\)_clip=0_direction=forward_netPrec=f32__inFmts=ncw\.ntc_outFmts=ncw\.ncw_primitive=ref_any)", // NOLINT
         // 98151. Not valid sorting for slices in reference.
         R"(.*UniqueLayerTestCPU.*axis.*True.*)",
+        // Issue: 104402. Incorrect broadcasting in FQ reference implentation
+        R"(.*smoke_FakeQuantizeLayerCPUTest_Decompos.*IS=\[4\.5\.6\.6\]_TS=\(\(4\.5\.6\.6\)\)_RS=\(\(1\.1\.6\.6\)\)_\(\(1\.1\.6\.6\)\)_\(\(1\.5\.6\.1\)\)_\(\(1\.5\.1\.6\)\).*)",
+        R"(.*smoke_FakeQuantizeLayerCPUTest_Decompos.*IS=\[4\.5\.6\.6\]_TS=\(\(4\.5\.6\.6\)\)_RS=\(\(1\.5\.6\.1\)\)_\(\(1\.5\.6\.1\)\)_\(\(1\.5\.6\.1\)\)_\(\(1\.5\.1\.6\)\).*)",
     };
 
 #define FIX_62820 0