diff --git a/NEWS.md b/NEWS.md
index f8efaa5f1f..9ef9fc92f8 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -2,7 +2,8 @@
 
 ## v0.12.0
 
-* The Dense layer now supports inputs with [multiple batch dimensions](https://github.com/FluxML/Flux.jl/pull/1405)
+* The Dense layer now supports inputs with [multiple batch dimensions](https://github.com/FluxML/Flux.jl/pull/1405).
+* Dense and Conv layers no longer perform  [implicit type conversion](https://github.com/FluxML/Flux.jl/pull/1394).
 * Excise datasets in favour of other providers in the julia ecosystem.
 * Added option to set `bias` to [false](https://github.com/FluxML/Flux.jl/pull/1379) to eliminating `bias` from being trained.
 * Removed kwarg only constructors for [`convolutional layers`](https://github.com/FluxML/Flux.jl/pull/1379).
diff --git a/docs/src/performance.md b/docs/src/performance.md
index c78ff1a115..36eea32e95 100644
--- a/docs/src/performance.md
+++ b/docs/src/performance.md
@@ -13,7 +13,6 @@ not because the operations are faster, but because the memory usage is halved.
 Which means allocations occur much faster.
 And you use less memory.
 
-
 ## Preserve inputs' types
 
 Not only should your activation and loss functions be [type-stable](https://docs.julialang.org/en/v1/manual/performance-tips/#Write-%22type-stable%22-functions-1),
@@ -21,8 +20,8 @@ they should also preserve the type of their inputs.
 
 A very artificial example using an activation function like
 
-```
-    my_tanh(x) = Float64(tanh(x))
+```julia
+my_tanh(x) = Float64(tanh(x))
 ```
 
 will result in performance on `Float32` input orders of magnitude slower than the normal `tanh` would,
@@ -35,20 +34,21 @@ you will see a large slow-down.
 This can occur sneakily, because you can cause type-promotion by interacting with a numeric literals.
 E.g. the following will have run into the same problem as above:
 
-```
-    leaky_tanh(x) = 0.01*x + tanh(x)
+```julia
+leaky_tanh(x) = 0.01*x + tanh(x)
 ```
 
 While one could change the activation function (e.g. to use `0.01f0*x`), the idiomatic (and safe way)  to avoid type casts whenever inputs changes is to use `oftype`:
-```
-    leaky_tanh(x) = oftype(x/1, 0.01)*x + tanh(x)
-```
 
+```julia
+leaky_tanh(x) = oftype(x/1, 0.01)*x + tanh(x)
+```
 
 ## Evaluate batches as Matrices of features
 
 While it can sometimes be tempting to process your observations (feature vectors) one at a time
 e.g.
+
 ```julia
 function loss_total(xs::AbstractVector{<:Vector}, ys::AbstractVector{<:Vector})
     sum(zip(xs, ys)) do (x, y_target)
diff --git a/src/layers/basic.jl b/src/layers/basic.jl
index 6a4be2dcab..c8d905be46 100644
--- a/src/layers/basic.jl
+++ b/src/layers/basic.jl
@@ -121,9 +121,8 @@ end
 
 function (a::Dense)(x::AbstractArray)
   W, b, σ = a.W, a.b, a.σ
-  # reshape to handle dims > 1 as batch dimensions
   sz = size(x)
-  x = reshape(x, sz[1], :) 
+  x = reshape(x, sz[1], :) # reshape to handle dims > 1 as batch dimensions 
   x = σ.(W*x .+ b)
   return reshape(x, :, sz[2:end]...)
 end
@@ -134,14 +133,6 @@ function Base.show(io::IO, l::Dense)
   print(io, ")")
 end
 
-# Try to avoid hitting generic matmul in some simple cases
-# Base's matmul is so slow that it's worth the extra conversion to hit BLAS
-(a::Dense{<:Any,W})(x::AbstractArray{T}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  invoke(a, Tuple{AbstractArray}, x)
-
-(a::Dense{<:Any,W})(x::AbstractArray{<:AbstractFloat}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  a(T.(x))
-
 """
     Diagonal(in::Integer)
 
diff --git a/src/layers/conv.jl b/src/layers/conv.jl
index b42f9bdae2..8724810762 100644
--- a/src/layers/conv.jl
+++ b/src/layers/conv.jl
@@ -164,11 +164,6 @@ function Base.show(io::IO, l::Conv)
   print(io, ")")
 end
 
-(a::Conv{<:Any,<:Any,W})(x::AbstractArray{T}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  invoke(a, Tuple{AbstractArray}, x)
-
-(a::Conv{<:Any,<:Any,W})(x::AbstractArray{<:Real}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  a(T.(x))
 
 """
     ConvTranspose(filter, in => out, σ=identity; stride=1, pad=0, dilation=1)
@@ -265,11 +260,6 @@ function Base.show(io::IO, l::ConvTranspose)
   print(io, ")")
 end
 
-(a::ConvTranspose{<:Any,<:Any,W})(x::AbstractArray{T}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  invoke(a, Tuple{AbstractArray}, x)
-
-(a::ConvTranspose{<:Any,<:Any,W})(x::AbstractArray{<:Real}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  a(T.(x))
 
 function calc_padding(::Type{ConvTranspose}, pad::SamePad, k::NTuple{N,T}, dilation, stride) where {N,T}
   calc_padding(Conv, pad, k .- stride .+ 1, dilation, stride)
@@ -363,11 +353,6 @@ function Base.show(io::IO, l::DepthwiseConv)
   print(io, ")")
 end
 
-(a::DepthwiseConv{<:Any,<:Any,W})(x::AbstractArray{T}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  invoke(a, Tuple{AbstractArray}, x)
-
-(a::DepthwiseConv{<:Any,<:Any,W})(x::AbstractArray{<:Real}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  a(T.(x))
 
 """
     CrossCor(filter, in => out, σ=identity; stride=1, pad=0, dilation=1)
@@ -449,12 +434,6 @@ function Base.show(io::IO, l::CrossCor)
   print(io, ")")
 end
 
-(a::CrossCor{<:Any,<:Any,W})(x::AbstractArray{T}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  invoke(a, Tuple{AbstractArray}, x)
-
-(a::CrossCor{<:Any,<:Any,W})(x::AbstractArray{<:Real}) where {T <: Union{Float32,Float64}, W <: AbstractArray{T}} =
-  a(T.(x))
-
 """
     AdaptiveMaxPool(out::NTuple)
 
diff --git a/test/utils.jl b/test/utils.jl
index 2d4537d871..377dcccc49 100644
--- a/test/utils.jl
+++ b/test/utils.jl
@@ -145,10 +145,13 @@ end
 
 @testset "Precision" begin
   m = Chain(Dense(10, 5, relu), Dense(5, 2))
-  x = rand(10)
+  x64 = rand(Float64, 10)
+  x32 = rand(Float32, 10)
   @test eltype(m[1].W) == Float32
-  @test eltype(m(x)) == Float32
-  @test eltype(f64(m)(x)) == Float64
+  @test eltype(m(x32)) == Float32
+  @test eltype(m(x64)) == Float64
+  @test eltype(f64(m)(x32)) == Float64
+  @test eltype(f64(m)(x64)) == Float64
   @test eltype(f64(m)[1].W) == Float64
   @test eltype(f32(f64(m))[1].W) == Float32
 end