diff --git a/src/ThreadsX.jl b/src/ThreadsX.jl
index 1840d681..1ea6a2d0 100644
--- a/src/ThreadsX.jl
+++ b/src/ThreadsX.jl
@@ -38,7 +38,8 @@ module Implementations
 import SplittablesBase
 using ArgCheck: @argcheck, @check
 using BangBang: SingletonVector, append!!, push!!, union!!
-using Base: Ordering, add_sum, mapreduce_empty, mul_prod, reduce_empty
+using Base:
+    HasShape, IteratorSize, Ordering, add_sum, mapreduce_empty, mul_prod, reduce_empty
 using ConstructionBase: setproperties
 using InitialValues: asmonoid
 using Referenceables: referenceable
diff --git a/src/basesizes.jl b/src/basesizes.jl
index fa4ec7d4..67d6fd13 100644
--- a/src/basesizes.jl
+++ b/src/basesizes.jl
@@ -1,3 +1,7 @@
+default_basesize(n::Integer) = max(1, cld(n, (8 * Threads.nthreads())))
+default_basesize(xs) =
+    default_basesize(SplittablesBase.amount(last(extract_transducer(xs))))
+
 default_basesize(_, _, xs) = default_basesize(xs::AbstractArray)
 
 # TODO: handle `Base.Fix2` etc.
diff --git a/src/map.jl b/src/map.jl
index 49a95eae..27b9b98a 100644
--- a/src/map.jl
+++ b/src/map.jl
@@ -3,10 +3,16 @@ __map(f, itr; kwargs...) =
 __map(f, itrs...; kwargs...) =
     tcollect(MapSplat(f), zip(itrs...); basesize = default_basesize(itrs[1]), kwargs...)
 
+reshape_as(ys, xs) = reshape_as(ys, xs, IteratorSize(xs))
+reshape_as(ys, xs, ::IteratorSize) = ys
+reshape_as(ys, xs, ::HasShape) = reshape(ys, size(xs))
+reshape_as(::Empty{T}, xs, isize::HasShape) where {T<:AbstractVector} =
+    reshape_as(T(undef, length(xs)), xs, isize)
+
 function _map(f, itr, itrs...; kwargs...)
     ys = __map(f, itr, itrs...; kwargs...)
     isempty(ys) && return map(f, itr, itrs...)
-    return ys
+    return reshape_as(ys, itr)
 end
 
 ThreadsX.map(f, itr, itrs...; kwargs...) = _map(f, itr, itrs...; kwargs...)
@@ -36,8 +42,10 @@ end
 struct ConvertTo{T} end
 (::ConvertTo{T})(x) where {T} = convert(T, x)
 
-ThreadsX.collect(::Type{T}, itr; kwargs...) where {T} =
-    tcopy(Map(ConvertTo{T}()), Vector{T}, itr; basesize = default_basesize(itr), kwargs...)
+ThreadsX.collect(::Type{T}, itr; kwargs...) where {T} = reshape_as(
+    tcopy(Map(ConvertTo{T}()), Vector{T}, itr; basesize = default_basesize(itr), kwargs...),
+    itr,
+)
 
 ThreadsX.collect(itr; kwargs...) =
-    tcollect(itr; basesize = default_basesize(itr), kwargs...)
+    reshape_as(tcollect(itr; basesize = default_basesize(itr), kwargs...), itr)
diff --git a/src/utils.jl b/src/utils.jl
index b0b60cf0..311e721e 100644
--- a/src/utils.jl
+++ b/src/utils.jl
@@ -1,6 +1,3 @@
-default_basesize(n::Integer) = max(1, cld(n, (8 * Threads.nthreads())))
-default_basesize(xs) = default_basesize(length(xs))
-
 function adhoc_partition(xs, n)
     @check firstindex(xs) == 1
     m = cld(length(xs), n)
diff --git a/test/test_with_base.jl b/test/test_with_base.jl
index 1c4f3bc3..d135dbb5 100644
--- a/test/test_with_base.jl
+++ b/test/test_with_base.jl
@@ -1,5 +1,6 @@
 module TestWithBase
 
+using Base: splat
 using Test
 using ThreadsX
 
@@ -10,9 +11,16 @@ inc(x) = x + 1
 raw_testdata = """
 collect(1:10)
 collect(Float64, 1:10)
+collect(x for x in 1:10 if isodd(x))
+collect(Float64, (x for x in 1:10 if isodd(x)))
 collect(inc(x) for x in 1:10)
 collect(Float64, (inc(x) for x in 1:10))
+collect(x * y for x in 1:10, y in 11:20)
+collect(Float64, (x * y for x in 1:10, y in 11:20))
+collect(x * y for x in 1:0, y in 11:20)
+collect(Float64, (x * y for x in 1:0, y in 11:20))
 map(inc, 1:10)
+map(inc, (x for x in 1:10 if isodd(x)))
 map(inc, Float64[])
 map(inc, ones(3, 3))
 map(inc, ones(3, 0))
@@ -21,6 +29,8 @@ map(*, 1:10, 11:20)
 map(*, ones(3, 3), ones(3, 3))
 map(*, ones(3, 0), ones(3, 0))
 map(*, ones(0, 3), ones(0, 3))
+map(splat(*), Iterators.product(1:10, 11:20))
+map(splat(*), Iterators.product(1:0, 11:20))
 reduce(+, 1:10)
 reduce(+, 1:0)
 reduce(+, Bool[])