diff --git a/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_support_checker.cc b/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_support_checker.cc
index aebba40590..cf0b555778 100644
--- a/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_support_checker.cc
+++ b/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_support_checker.cc
@@ -1566,10 +1566,23 @@ class UnaryOpSupportChecker : public BaseOpSupportChecker {
 
 bool UnaryOpSupportChecker::IsOpSupportedImpl(const InitializedTensorSet& initializers, const NodeUnit& node_unit,
                                               const OpSupportCheckParams& params) const {
-  if (node_unit.OpType() == "QLinearSigmoid")
+  if (node_unit.OpType() == "QLinearSigmoid") {
     return IsQuantizedOpSupported(initializers, node_unit, params);
-  else  // Everything except "QLinearSigmoid" are by default supported
+  } else if (node_unit.OpType() == "Sigmoid") {
+    Shape input_shape;
+    if (!GetShape(node_unit.Inputs()[0].node_arg, input_shape))
+      return false;
+
+    const auto input_size = input_shape.size();
+    if (input_size > 4 || input_size == 0) {
+      LOGS_DEFAULT(VERBOSE) << "ANEURALNETWORKS_LOGISTIC only supports 1-4d shape, input is "
+                            << input_size << "d shape";
+      return false;
+    }
     return true;
+  }
+  // Everything else are by default supported
+  return true;
 }
 
 int32_t UnaryOpSupportChecker::GetMinSupportedNNAPIFeatureLevel(const NodeUnit& node_unit,
diff --git a/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc b/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc
index e16c4b15f8..4f372ec833 100644
--- a/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc
+++ b/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc
@@ -85,6 +85,30 @@ TEST(NnapiExecutionProviderTest, ReshapeFlattenTest) {
 #endif
 }
 
+TEST(NnapiExecutionProviderTest, SigmoidSupportedInputRankTest) {
+  const ORTCHAR_T* model_file_name = ORT_TSTR("testdata/nnapi_sigmoid_input_rank_test.onnx");
+
+#if defined(__ANDROID__)
+  std::vector<int64_t> dims_mul_x = {2, 1, 2, 1, 2};
+  std::vector<float> values_mul_x = {1.0f, 2.0f, 3.0f, 4.0f, 1.0f, 2.0f, 3.0f, 4.0f};
+
+  OrtValue ml_value_x;
+  CreateMLValue<float>(TestNnapiExecutionProvider()->GetAllocator(0, OrtMemTypeDefault), dims_mul_x, values_mul_x,
+                       &ml_value_x);
+  NameMLValMap feeds;
+  feeds.insert(std::make_pair("X", ml_value_x));
+
+  RunAndVerifyOutputsWithEP(model_file_name, "NnapiExecutionProviderTest.SigmoidSupportedInputRankTest",
+                            std::make_unique<NnapiExecutionProvider>(0),
+                            feeds, {ExpectedEPNodeAssignment::None} /* params */);
+#else
+  // test load only
+  TestModelLoad(model_file_name,
+                [](const Graph& graph) { ASSERT_EQ(CountAssignedNodes(graph, kNnapiExecutionProvider), 0)
+                                             << "No nodes should have been taken by the NNAPI EP"; });
+#endif
+}
+
 // Since NNAPI EP does not support dynamic shape input and we now switch from the approach of immediately rejecting
 // the whole graph in NNAPI EP if it has a dynamic input to check at individual operator support check level, we have a
 // separated test here.
diff --git a/onnxruntime/test/testdata/nnapi_sigmoid_input_rank_test.onnx b/onnxruntime/test/testdata/nnapi_sigmoid_input_rank_test.onnx
new file mode 100644
index 0000000000..fef9e610bc
--- /dev/null
+++ b/onnxruntime/test/testdata/nnapi_sigmoid_input_rank_test.onnx
@@ -0,0 +1,17 @@
+:{
+
+
X
Ysigmoid"SigmoidNnapi_sigmoid_input_rank_testZ
+
X
+

+
+

+
+

+b
+
Y
+

+
+

+
+

+B
\ No newline at end of file
diff --git a/onnxruntime/test/testdata/nnapi_sigmoid_input_rank_test.py b/onnxruntime/test/testdata/nnapi_sigmoid_input_rank_test.py
new file mode 100644
index 0000000000..160e8ba73a
--- /dev/null
+++ b/onnxruntime/test/testdata/nnapi_sigmoid_input_rank_test.py
@@ -0,0 +1,25 @@
+import onnx
+from onnx import TensorProto, helper
+
+# Since NNAPI ANEURALNETWORKS_LOGISTIC only supports input tensor rank up to 4, we added limits in
+# Sigmoid op support checker in NNAPI EP, so we don't fail hard. Added test case here.
+
+
+def GenerateModel(model_name):
+    node = [
+        helper.make_node("Sigmoid", ["X"], ["Y"], "sigmoid"),
+    ]
+
+    graph = helper.make_graph(
+        node,
+        "Nnapi_sigmoid_input_rank_test",
+        [helper.make_tensor_value_info("X", TensorProto.FLOAT, [2, 1, 2, 1, 2])],  # input
+        [helper.make_tensor_value_info("Y", TensorProto.FLOAT, [2, 1, 2, 1, 2])],  # output
+    )
+
+    model = helper.make_model(graph)
+    onnx.save(model, model_name)
+
+
+if __name__ == "__main__":
+    GenerateModel("nnapi_sigmoid_input_rank_test.onnx")