diff --git a/lib/async/scheduler.rb b/lib/async/scheduler.rb
index f128896d..ffdd9e17 100644
--- a/lib/async/scheduler.rb
+++ b/lib/async/scheduler.rb
@@ -307,6 +307,7 @@ def address_resolve(hostname)
 		# @parameter io [IO] The IO object to wait on.
 		# @parameter events [Integer] The events to wait for, e.g. `IO::READABLE`, `IO::WRITABLE`, etc.
 		# @parameter timeout [Float | Nil] The maximum time to wait, or if nil, indefinitely.
+		# @returns [Integer | false] The subset of events that are ready, or `false` if the timeout expires.
 		def io_wait(io, events, timeout = nil)
 			fiber = Fiber.current
 			expired = false
@@ -326,8 +327,8 @@ def io_wait(io, events, timeout = nil)
 			
 			# A selector wait may return a falsy result when the fiber is resumed without the requested IO becoming ready. For example, a deferred unblock from a previous blocking operation may arrive after the fiber has moved on to this wait. Retry these stale or spurious wake-ups without resetting the original timer.
 			until result = @selector.io_wait(fiber, io, events)
-				# If the original timer resumed the fiber, the falsy result represents the timeout rather than a spurious wake-up:
-				return nil if expired
+				# If the original timer resumed the fiber, the false result represents the timeout rather than a spurious wake-up:
+				return false if expired
 			end
 			
 			return result
diff --git a/releases.md b/releases.md
index dc0ff4bf..158857c0 100644
--- a/releases.md
+++ b/releases.md
@@ -1,5 +1,9 @@
 # Releases
 
+## Unreleased
+
+  - Fixed `Scheduler#io_wait` returning `nil` instead of `false` when an explicit timeout expired. Native callers such as `Socket#connect` with `connect_timeout:` distinguish a timeout by checking for `false`, so the `nil` caused `TypeError: no implicit conversion from nil to integer` instead of the intended `IO::TimeoutError`.
+
 ## v2.45.0
 
   - Fixed scheduler I/O and process waits returning prematurely after stale or interrupted wake-ups. I/O waits now preserve their original timeout, while blocking process waits retry and non-blocking `Process::WNOHANG` waits still return `nil`.
diff --git a/test/async/scheduler/io.rb b/test/async/scheduler/io.rb
index 3a7389dd..cc2aa8fa 100644
--- a/test/async/scheduler/io.rb
+++ b/test/async/scheduler/io.rb
@@ -24,6 +24,18 @@
 			s2.close
 		end
 		
+		it "returns false from io_wait when an explicit timeout expires" do
+			s1, s2 = Socket.pair :UNIX, :STREAM, 0
+			
+			# Native callers (e.g. `wait_connectable` in `Socket#connect` with `connect_timeout:`) distinguish a timeout from a readiness mask by checking for `false`, matching the non-scheduler `rb_io_wait`:
+			result = reactor.io_wait(s1, IO::READABLE, 0.001)
+			
+			expect(result).to be == false
+		ensure
+			s1.close
+			s2.close
+		end
+		
 		it "can read a single character" do
 			s1, s2 = Socket.pair :UNIX, :STREAM, 0