diff --git a/src/host/abstractarray.jl b/src/host/abstractarray.jl
index 2016f2cb..c05e8b19 100644
--- a/src/host/abstractarray.jl
+++ b/src/host/abstractarray.jl
@@ -350,6 +350,18 @@ function Base.copyto!(dest::AnyGPUArray{T}, dstart::Integer,
     return dest
 end
 
+# `BitArray`s pack their elements into bits, so unpack them on the CPU too
+function Base.copyto!(dest::AnyGPUArray{T}, dstart::Integer,
+                      src::BitArray, sstart::Integer, n::Integer) where {T}
+    n == 0 && return dest
+    temp = Vector{T}(undef, n)
+    copyto!(temp, 1, src, sstart, n)
+    copyto!(dest, dstart, temp, 1, n)
+    return dest
+end
+
+Base.copyto!(dest::AnyGPUArray, src::BitArray) = copyto!(dest, 1, src, 1, length(src))
+
 ## generalized blocks of heterogeneous memory
 
 # mirrors the checks of Base's `copyto!(dest, Rdest, src, Rsrc)`
diff --git a/test/testsuite/base.jl b/test/testsuite/base.jl
index 55e2bac1..832b1be4 100644
--- a/test/testsuite/base.jl
+++ b/test/testsuite/base.jl
@@ -210,6 +210,29 @@ end
                 @test Array(src) == dst
             end
         end
+
+        # from a `BitArray`, without scalar indexing (#836)
+        let src = BitVector([true, false, true, true, false])
+            dst = AT{Bool}(undef, length(src))
+            copyto!(dst, src)
+            @test Array(dst) == src
+
+            dst = AT(fill(false, 7))
+            copyto!(dst, 2, src, 3, 2)
+            @test Array(dst) == [false, true, true, false, false, false, false]
+
+            src = BitMatrix([true false; false true])
+            dst = AT{Bool}(undef, 2, 2)
+            copyto!(dst, src)
+            @test Array(dst) == src
+        end
+        if Float32 in eltypes
+            let src = BitVector([true, false, true])
+                dst = AT{Float32}(undef, 3)
+                copyto!(dst, src)
+                @test Array(dst) == Float32[1, 0, 1]
+            end
+        end
     end
 
     @testset "copyto! of blocks" begin