diff --git a/src/layers/conv.jl b/src/layers/conv.jl
index 338bb89725..f7c4ed2498 100644
--- a/src/layers/conv.jl
+++ b/src/layers/conv.jl
@@ -44,8 +44,25 @@ julia> layer3(xs) |> size  # output size = `ceil(input_size/stride)` = 50
 ```
 """
 struct SamePad end
-
-calc_padding(lt, pad, k::NTuple{N,T}, dilation, stride) where {T,N} = expand(Val(2*N), pad)
+calc_padding(lt, pad::Int, k::NTuple{N,T}, dilation, stride) where {T,N} = expand(Val(2*N), pad)
+function calc_padding(lt, pad::NTuple{Np, T}, k::NTuple{Nk, T}, dilation, stride) where {T, Nk, Np}
+  # calc_padding for when a tuple is passed as padding.
+  if  Nk == Np
+    # duplicate each dim
+    new_pad = []
+    for i in pad
+      push!(new_pad, i)
+      push!(new_pad, i)
+    end
+    return tuple(new_pad...)
+  elseif Nk == 2Np
+    # Copy as it is
+    return pad
+  else
+    # Error out
+    throw(ArgumentError("invalid padding dimensions"))
+  end
+end
 function calc_padding(lt, ::SamePad, k::NTuple{N,T}, dilation, stride) where {N,T}
   #Ref: "A guide to convolution arithmetic for deep learning" https://arxiv.org/abs/1603.07285
 
diff --git a/test/layers/conv.jl b/test/layers/conv.jl
index be26786495..f3dd6e8d3a 100644
--- a/test/layers/conv.jl
+++ b/test/layers/conv.jl
@@ -208,6 +208,16 @@ end
   m1 = ConvTranspose((3,5,3), 3=>6, stride=3)
   m2 = ConvTranspose((3,5,3), 3=>6, stride=3, outpad=(1,0,1))
   @test size(m2(x))[1:3] == (size(m1(x))[1:3] .+ (1,0,1))
+
+  # test ConvTranspose constructor with tuple padding
+  kernel_dims = (3, 3)
+  pad = (1, 0)
+  x = randn(Float32, 5, 6, 1, 16)
+  m = ConvTranspose(kernel_dims, 1=>1, pad=pad)
+  result = m(x)
+  # Formula obtained from https://makeyourownneuralnetwork.blogspot.com/2020/02/calculating-output-size-of-convolutions.html
+  output_dims(pad_dims, kernel_dim, input_dim, stride) = (input_dim.-1).*stride.-(pad_dims.*2).+(kernel_dim.-1).+1
+  @test output_dims(m.pad, kernel_dims, size(x)[1:2], m.stride) == size(result)[1:2]
 end
 
 @testset "CrossCor" begin