diff --git a/TitanEngine/Global.Debugger.cpp b/TitanEngine/Global.Debugger.cpp
index f267967..11e4c5f 100644
--- a/TitanEngine/Global.Debugger.cpp
+++ b/TitanEngine/Global.Debugger.cpp
@@ -29,9 +29,7 @@ DWORD ProcessExitCode = 0;
 HANDLE DBGFileHandle;
 std::vector<ULONG_PTR> tlsCallBackList;
 std::vector<PROCESS_ITEM_DATA> hListProcess;
-DWORD engineStepCount = 0;
-LPVOID engineStepCallBack = NULL;
-bool engineStepActive = false;
+std::unordered_map<DWORD, EngineStepThreadState> engineStepThreads;
 bool engineProcessIsNowDetached = false;
 DWORD DBGCode = DBG_CONTINUE;
 bool engineFileIsBeingDebugged = false;
@@ -90,6 +88,7 @@ void DebuggerReset()
     std::vector<BreakPointDetail>().swap(BreakPointBuffer);
     std::unordered_map<ULONG_PTR, MemoryBreakpointPageDetail>().swap(MemoryBreakpointPages);
     recentlyDeletedBpx.clear();
+    engineStepThreads.clear();
 }
 
 void ClearProcessList()
diff --git a/TitanEngine/Global.Debugger.h b/TitanEngine/Global.Debugger.h
index 1861756..d0b8e11 100644
--- a/TitanEngine/Global.Debugger.h
+++ b/TitanEngine/Global.Debugger.h
@@ -2,6 +2,7 @@
 #define _GLOBAL_DEBUGGER_H
 
 #include <vector>
+#include <unordered_map>
 #include <Windows.h>
 
 extern HARDWARE_DATA DebugRegister[4];
@@ -29,9 +30,18 @@ extern DWORD ProcessExitCode;
 extern HANDLE DBGFileHandle;
 extern std::vector<ULONG_PTR> tlsCallBackList;
 extern std::vector<PROCESS_ITEM_DATA> hListProcess;
-extern DWORD engineStepCount;
-extern LPVOID engineStepCallBack;
-extern bool engineStepActive;
+// Per-thread pending single-step state. A single-step (user step or the internal
+// step-over of a breakpoint) is always armed on a specific thread by setting that
+// thread's trap flag, but the resulting STATUS_SINGLE_STEP event can be delivered
+// interleaved with other threads' events. Keying the step state by thread id makes
+// each thread consume only its own step, and allows several threads to be stepped
+// independently (matching GleeBug).
+struct EngineStepThreadState
+{
+    LPVOID callback; // step callback to fire when the step completes
+    DWORD count;     // remaining repeats (0 = fire the callback on the next step)
+};
+extern std::unordered_map<DWORD, EngineStepThreadState> engineStepThreads;
 extern bool engineProcessIsNowDetached;
 extern DWORD DBGCode;
 extern bool engineFileIsBeingDebugged;
diff --git a/TitanEngine/TitanEngine.Debugger.Control.cpp b/TitanEngine/TitanEngine.Debugger.Control.cpp
index 548f746..3984c0d 100644
--- a/TitanEngine/TitanEngine.Debugger.Control.cpp
+++ b/TitanEngine/TitanEngine.Debugger.Control.cpp
@@ -38,7 +38,9 @@ __declspec(dllexport) void TITCALL ForceClose()
 __declspec(dllexport) void TITCALL StepInto(LPVOID StepCallBack)
 {
     EnterCriticalSection(&engineStepActiveCr);
-    if(!engineStepActive)
+    // Arm a single-step for the current event thread only. State is per-thread, so a
+    // step already pending on another thread does not block this one.
+    if(engineStepThreads.find(DBGEvent.dwThreadId) == engineStepThreads.end())
     {
         ULONG_PTR ueCurrentPosition = GetContextData(UE_CIP);
         unsigned char instr[16];
@@ -60,9 +62,7 @@ __declspec(dllexport) void TITCALL StepInto(LPVOID StepCallBack)
             myDBGContext.EFlags |= UE_TRAP_FLAG;
             SetThreadContext(hActiveThread, &myDBGContext);
             EngineCloseHandle(hActiveThread);
-            engineStepActive = true;
-            engineStepCallBack = StepCallBack;
-            engineStepCount = 0;
+            engineStepThreads[DBGEvent.dwThreadId] = { StepCallBack, 0 };
         }
     }
     LeaveCriticalSection(&engineStepActiveCr);
@@ -93,7 +93,11 @@ __declspec(dllexport) void TITCALL StepOut(LPVOID StepOut, bool StepFinal)
 __declspec(dllexport) void TITCALL SingleStep(DWORD StepCount, LPVOID StepCallBack)
 {
     StepInto(StepCallBack);
-    engineStepCount = StepCount - 1; //We already stepped once
+    EnterCriticalSection(&engineStepActiveCr);
+    auto it = engineStepThreads.find(DBGEvent.dwThreadId);
+    if(it != engineStepThreads.end())
+        it->second.count = StepCount - 1; //We already stepped once
+    LeaveCriticalSection(&engineStepActiveCr);
 }
 
 __declspec(dllexport) void TITCALL SetNextDbgContinueStatus(DWORD SetDbgCode)
diff --git a/TitanEngine/TitanEngine.Debugger.DebugLoop.cpp b/TitanEngine/TitanEngine.Debugger.DebugLoop.cpp
index d11889f..4641709 100644
--- a/TitanEngine/TitanEngine.Debugger.DebugLoop.cpp
+++ b/TitanEngine/TitanEngine.Debugger.DebugLoop.cpp
@@ -16,22 +16,28 @@
 static void engineStep()
 {
     EnterCriticalSection(&engineStepActiveCr);
-    if(engineStepActive)
+    // Only deliver the pending single-step to the thread it was armed for. State is
+    // per-thread, so an automatic breakpoint step-over on a different thread cannot
+    // consume this thread's step or fire its callback on the wrong thread.
+    auto it = engineStepThreads.find(DBGEvent.dwThreadId);
+    if(it != engineStepThreads.end())
     {
         DBGCode = DBG_CONTINUE;
-        if(engineStepCount == 0)
+        if(it->second.count == 0)
         {
             typedef void(TITCALL * fCustomBreakPoint)(void);
-            auto cbStep = fCustomBreakPoint(engineStepCallBack);
-            engineStepActive = false;
-            engineStepCallBack = NULL;
+            auto cbStep = fCustomBreakPoint(it->second.callback);
+            engineStepThreads.erase(it);
             LeaveCriticalSection(&engineStepActiveCr);
             cbStep();
         }
         else
         {
-            SingleStep(engineStepCount, engineStepCallBack);
+            auto count = it->second.count;
+            auto callback = it->second.callback;
+            engineStepThreads.erase(it); // SingleStep re-arms this thread's entry
             LeaveCriticalSection(&engineStepActiveCr);
+            SingleStep(count, callback);
         }
     }
     else
@@ -1288,7 +1294,11 @@ __declspec(dllexport) void TITCALL DebugLoop()
             //general unhandled exception callback
             if(DBGCode == DBG_EXCEPTION_NOT_HANDLED)
             {
-                engineStepActive = false;
+                // The current thread's pending single-step (if any) did not complete
+                // normally; cancel it. Other threads' steps are unaffected.
+                EnterCriticalSection(&engineStepActiveCr);
+                engineStepThreads.erase(DBGEvent.dwThreadId);
+                LeaveCriticalSection(&engineStepActiveCr);
 
                 if(DBGCustomHandler->chUnhandledException != NULL)
                 {