diff --git a/src/decompiler.cc b/src/decompiler.cc
index bd6a5af732..fc6f4898f3 100644
--- a/src/decompiler.cc
+++ b/src/decompiler.cc
@@ -226,16 +226,62 @@ struct Decompiler {
 
   void LoadStore(Value &val, const Node& addr_exp, uint32_t offset,
                   Opcode opc, Address align, Type op_type) {
-    BracketIfNeeded(val, Precedence::Indexing);
     auto access = lst.GenAccess(offset, addr_exp);
+    if (!access.empty()) {
+      // We can do this load/store as a struct access.
+      BracketIfNeeded(val, Precedence::Indexing);
+      val.v.back() += "." + access;
+      return;
+    }
+    // Do the load/store as a generalized indexing operation.
+    // The offset is divisible by the alignment in 99.99% of
+    // cases, but the spec doesn't guarantee it, so we must
+    // have a backup syntax.
+    auto index = offset % align == 0
+                ? std::to_string(offset / align)
+                : cat(std::to_string(offset), "@", std::to_string(align));
+    // Detect the very common case of (base + (index << 2))[0]:int etc.
+    // so we can instead do base[index]:int
+    // TODO: (index << 2) on the left of + occurs also.
+    // TODO: sadly this does not address cases where the shift amount > align.
+    // (which happens for arrays of structs or arrays of long (with align=4)).
+    // TODO: also very common is (v = base + (index << 2))[0]:int
+    if (addr_exp.etype == ExprType::Binary) {
+      auto& pe = *cast<BinaryExpr>(addr_exp.e);
+      auto& shift_exp = addr_exp.children[1];
+      if (pe.opcode == Opcode::I32Add &&
+          shift_exp.etype == ExprType::Binary) {
+        auto& se = *cast<BinaryExpr>(shift_exp.e);
+        auto& const_exp = shift_exp.children[1];
+        if (se.opcode == Opcode::I32Shl &&
+            const_exp.etype == ExprType::Const) {
+          auto& ce = *cast<ConstExpr>(const_exp.e);
+          if (ce.const_.type == Type::I32 && (1 << ce.const_.u32) == align) {
+            // Pfew, case detected :( Lets re-write this in Haskell.
+            // TODO: we're decompiling these twice.
+            // The thing to the left of << is going to be part of the index.
+            auto ival = DecompileExpr(shift_exp.children[0], &shift_exp);
+            if (ival.v.size() == 1) {  // Don't bother if huge.
+              if (offset == 0) {
+                index = ival.v[0];
+              } else {
+                BracketIfNeeded(ival, Precedence::Add);
+                index = cat(ival.v[0], " + ", index);
+              }
+              // We're going to use the thing to the left of + as the new
+              // base address:
+              val = DecompileExpr(addr_exp.children[0], &addr_exp);
+            }
+          }
+        }
+      }
+    }
+    BracketIfNeeded(val, Precedence::Indexing);
     val.v.back() +=
-        !access.empty()
-            ? "." + access
-            : cat("[", std::to_string(offset),
-                  "]:", GetDecompTypeName(GetMemoryType(op_type, opc)),
-                  !opc.IsNaturallyAligned(align)
-                      ? cat("@" + std::to_string(align))
-                      : "");
+        cat("[", index, "]:", GetDecompTypeName(GetMemoryType(op_type, opc)),
+            opc.IsNaturallyAligned(align)
+              ? ""
+              : cat("@", std::to_string(align)));
   }
 
   Value DecompileExpr(const Node& n, const Node* parent) {
diff --git a/test/decompile/basic.txt b/test/decompile/basic.txt
index 24d2a38520..34c6f57810 100644
--- a/test/decompile/basic.txt
+++ b/test/decompile/basic.txt
@@ -28,7 +28,7 @@
     if
       i32.const 1
       i32.const 2
-      i32.load offset=3
+      i32.load offset=3 align=1
       i32.const 5
       i32.add
       i32.store offset=4
@@ -126,7 +126,7 @@ export function f(a:int, b:int):int {
   var c:long = 8L;
   var d:float = 6.0f;
   var e:double = 7.0;
-  if (e < 10.0) { 1[4]:int = 2[3]:int + 5 }
+  if (e < 10.0) { 1[1]:int = 2[3]:int@1 + 5 }
   f(a + g_b, 9);
   loop L_b {
     if (if (0) { 1 } else { 2 }) goto B_c;
diff --git a/test/decompile/loadstore.txt b/test/decompile/loadstore.txt
index d1216950f2..a05d0e29c0 100644
--- a/test/decompile/loadstore.txt
+++ b/test/decompile/loadstore.txt
@@ -53,8 +53,48 @@
     drop
     ;; 4) Unaligned access / access with unexpected gaps.
     get_local 7
-    f32.load offset=1
+    f32.load offset=1 align=1
     drop
+    ;; Test index rewriting.
+    ;; code that does (base + (index << 2))[0]:int is super common.
+    get_local 0
+    get_local 1
+    i32.const 2
+    i32.shl
+    i32.add
+    get_local 0
+    get_local 1
+    i32.const 2
+    i32.shl
+    i32.add
+    i32.load offset=0
+    i32.store offset=0
+    ;; Same with non-zero offsets.
+    get_local 0
+    get_local 1
+    i32.const 2
+    i32.shl
+    i32.add
+    get_local 0
+    get_local 1
+    i32.const 2
+    i32.shl
+    i32.add
+    i32.load offset=4
+    i32.store offset=4
+    ;; If the shift amount does not match, it doesn't work.
+    get_local 0
+    get_local 1
+    i32.const 3
+    i32.shl
+    i32.add
+    get_local 0
+    get_local 1
+    i32.const 2
+    i32.shl
+    i32.add
+    i32.load offset=4
+    i32.store offset=4
   )
   (export "f" (func $f))
 )
@@ -77,7 +117,10 @@ export function f(a:{ a:float, b:float }, b:{ a:ushort, b:long }) {
   f[0]:short;
   g[0]:int;
   g[0]:int@1;
-  h[1]:float;
+  h[1]:float@1;
+  a[b]:int = a[b]:int;
+  a[b + 1]:int = a[b + 1]:int;
+  (a + (b << 3))[1]:int = a[b + 1]:int;
 }
 
 ;;; STDOUT ;;)