diff --git a/src/backends/aclCommon/ArmComputeUtils.hpp b/src/backends/aclCommon/ArmComputeUtils.hpp
index 0f56160..4d69090 100644
--- a/src/backends/aclCommon/ArmComputeUtils.hpp
+++ b/src/backends/aclCommon/ArmComputeUtils.hpp
@@ -135,9 +135,15 @@
     }
 }
 
-inline unsigned int ComputeSoftmaxAclAxis(const armnn::TensorInfo& tensor)
+inline unsigned int ComputeSoftmaxAclAxis(const SoftmaxDescriptor& softmaxDesc, const armnn::TensorInfo& tensor)
 {
-    unsigned int dim = tensor.GetNumDimensions();
+    // Detect the Android default value of -1 and return the ACL default value of 1.
+    if (softmaxDesc.m_Axis == -1)
+    {
+        return 1;
+    }
+
+   unsigned int dim = tensor.GetNumDimensions();
 
     BOOST_ASSERT(dim != 0);
 
diff --git a/src/backends/cl/workloads/ClSoftmaxBaseWorkload.cpp b/src/backends/cl/workloads/ClSoftmaxBaseWorkload.cpp
index 2f6d380..a355ba0 100644
--- a/src/backends/cl/workloads/ClSoftmaxBaseWorkload.cpp
+++ b/src/backends/cl/workloads/ClSoftmaxBaseWorkload.cpp
@@ -20,7 +20,7 @@
     const arm_compute::TensorInfo aclInputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(input);
     const arm_compute::TensorInfo aclOutputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(output);
 
-    unsigned int aclAxis = ComputeSoftmaxAclAxis(input);
+    unsigned int aclAxis = ComputeSoftmaxAclAxis(descriptor, input);
     return arm_compute::CLSoftmaxLayer::validate(&aclInputInfo, &aclOutputInfo, descriptor.m_Beta, aclAxis);
 }
 
diff --git a/src/backends/cl/workloads/ClSoftmaxFloatWorkload.cpp b/src/backends/cl/workloads/ClSoftmaxFloatWorkload.cpp
index f2f8d17..adb4872 100644
--- a/src/backends/cl/workloads/ClSoftmaxFloatWorkload.cpp
+++ b/src/backends/cl/workloads/ClSoftmaxFloatWorkload.cpp
@@ -23,7 +23,7 @@
     arm_compute::ICLTensor& input  = static_cast<ClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
     arm_compute::ICLTensor& output = static_cast<ClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
 
-    unsigned int aclAxis = ComputeSoftmaxAclAxis(info.m_InputTensorInfos[0]);
+    unsigned int aclAxis = ComputeSoftmaxAclAxis(m_Data.m_Parameters, info.m_InputTensorInfos[0]);
     m_SoftmaxLayer.configure(&input, &output, m_Data.m_Parameters.m_Beta, aclAxis);
 }
 
diff --git a/src/backends/cl/workloads/ClSoftmaxUint8Workload.cpp b/src/backends/cl/workloads/ClSoftmaxUint8Workload.cpp
index ce2a9e6..f14ea11 100644
--- a/src/backends/cl/workloads/ClSoftmaxUint8Workload.cpp
+++ b/src/backends/cl/workloads/ClSoftmaxUint8Workload.cpp
@@ -33,7 +33,7 @@
             "Invalid quantization for output. Only scale = 1.0f / 256.0f and offset = 0 supported");
     }
 
-    unsigned int aclAxis = ComputeSoftmaxAclAxis(info.m_InputTensorInfos[0]);
+    unsigned int aclAxis = ComputeSoftmaxAclAxis(m_Data.m_Parameters, info.m_InputTensorInfos[0]);
     m_SoftmaxLayer.configure(&input, &output, descriptor.m_Parameters.m_Beta, aclAxis);
 }
 
diff --git a/src/backends/neon/workloads/NeonSoftmaxBaseWorkload.cpp b/src/backends/neon/workloads/NeonSoftmaxBaseWorkload.cpp
index 8acb775..41ebfb9 100644
--- a/src/backends/neon/workloads/NeonSoftmaxBaseWorkload.cpp
+++ b/src/backends/neon/workloads/NeonSoftmaxBaseWorkload.cpp
@@ -20,7 +20,7 @@
     const arm_compute::TensorInfo aclInputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(input);
     const arm_compute::TensorInfo aclOutputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(output);
 
-    unsigned int aclAxis = ComputeSoftmaxAclAxis(input);
+    unsigned int aclAxis = ComputeSoftmaxAclAxis(descriptor, input);
     return arm_compute::NESoftmaxLayer::validate(&aclInputInfo, &aclOutputInfo, descriptor.m_Beta, aclAxis);
 }
 
diff --git a/src/backends/neon/workloads/NeonSoftmaxFloatWorkload.cpp b/src/backends/neon/workloads/NeonSoftmaxFloatWorkload.cpp
index 4dc9131..152d19c 100644
--- a/src/backends/neon/workloads/NeonSoftmaxFloatWorkload.cpp
+++ b/src/backends/neon/workloads/NeonSoftmaxFloatWorkload.cpp
@@ -24,7 +24,7 @@
     arm_compute::ITensor& output = boost::polymorphic_downcast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
 
     auto layer = std::make_unique<arm_compute::NESoftmaxLayer>(memoryManager);
-    unsigned int aclAxis = ComputeSoftmaxAclAxis(info.m_InputTensorInfos[0]);
+    unsigned int aclAxis = ComputeSoftmaxAclAxis(m_Data.m_Parameters, info.m_InputTensorInfos[0]);
     layer->configure(&input, &output, m_Data.m_Parameters.m_Beta, aclAxis);
     m_SoftmaxLayer.reset(layer.release());
 }
diff --git a/src/backends/neon/workloads/NeonSoftmaxUint8Workload.cpp b/src/backends/neon/workloads/NeonSoftmaxUint8Workload.cpp
index 8e83914..15a7066 100644
--- a/src/backends/neon/workloads/NeonSoftmaxUint8Workload.cpp
+++ b/src/backends/neon/workloads/NeonSoftmaxUint8Workload.cpp
@@ -34,7 +34,7 @@
     }
 
     auto layer = std::make_unique<arm_compute::NESoftmaxLayer>(memoryManager);
-    unsigned int aclAxis = ComputeSoftmaxAclAxis(info.m_InputTensorInfos[0]);
+    unsigned int aclAxis = ComputeSoftmaxAclAxis(m_Data.m_Parameters, info.m_InputTensorInfos[0]);
     layer->configure(&input, &output, descriptor.m_Parameters.m_Beta, aclAxis);
     m_SoftmaxLayer.reset(layer.release());
 }