diff --git a/src/backends/sdl/input.c b/src/backends/sdl/input.c
index 48f7774a..641bff9b 100644
--- a/src/backends/sdl/input.c
+++ b/src/backends/sdl/input.c
@@ -385,30 +385,40 @@ IN_Update(void)
 				}
 #endif
 
-				if ((event.key.keysym.sym >= SDLK_SPACE) &&
-					 (event.key.keysym.sym < SDLK_DELETE))
-				{
-					Key_Event(event.key.keysym.sym, true, false);
-				}
-				else
-				{
-					Key_Event(IN_TranslateSDLtoQ2Key(event.key.keysym.sym), true, true);
-				}
-
-				break;
-
+				// fall-through
 			case SDL_KEYUP:
+			{
+				qboolean down = (event.type == SDL_KEYDOWN);
 
-				if ((event.key.keysym.sym >= SDLK_SPACE) &&
-					 (event.key.keysym.sym < SDLK_DELETE))
+#if SDL_VERSION_ATLEAST(2, 0, 0)
+				// workaround for AZERTY-keyboards, which don't have 1, 2, ..., 9, 0 in first row:
+				// always map those physical keys (scancodes) to those keycodes anyway
+				// see also https://bugzilla.libsdl.org/show_bug.cgi?id=3188
+				SDL_Scancode sc = event.key.keysym.scancode;
+				if(sc >= SDL_SCANCODE_1 && sc <= SDL_SCANCODE_0)
 				{
-					Key_Event(event.key.keysym.sym, false, false);
+					// Note that the SDL_SCANCODEs are SDL_SCANCODE_1, _2, ..., _9, SDL_SCANCODE_0
+					// while in ASCII it's '0', '1', ..., '9' => handle 0 and 1-9 separately
+					// (quake2 uses the ASCII values for those keys)
+					int key = '0'; // implicitly handles SDL_SCANCODE_0
+					if(sc >= SDL_SCANCODE_1 && sc <= SDL_SCANCODE_9)
+					{
+						key = '1' + (sc - SDL_SCANCODE_1);
+					}
+					Key_Event(key, down, false);
+				}
+				else
+#endif // SDL2; SDL1.2 doesn't have scancodes
+				   if ((event.key.keysym.sym >= SDLK_SPACE) &&
+				       (event.key.keysym.sym < SDLK_DELETE))
+				{
+					Key_Event(event.key.keysym.sym, down, false);
 				}
 				else
 				{
-					Key_Event(IN_TranslateSDLtoQ2Key(event.key.keysym.sym), false, true);
+					Key_Event(IN_TranslateSDLtoQ2Key(event.key.keysym.sym), down, true);
 				}
-
+			}
 				break;
 
 #if SDL_VERSION_ATLEAST(2, 0, 0)