diff --git a/src/GPU3D.cpp b/src/GPU3D.cpp
index 4bb56ee4..9179d469 100644
--- a/src/GPU3D.cpp
+++ b/src/GPU3D.cpp
@@ -751,11 +751,6 @@ void SubmitPolygon()
     poly->Translucent = ((texfmt == 1 || texfmt == 6) && !(CurPolygonAttr & 0x10)) || (polyalpha > 0 && polyalpha < 31);
 
     poly->IsShadowMask = ((CurPolygonAttr & 0x3F000030) == 0x00000030);
-    if ((NumPolygons == 1) || (!CurPolygonRAM[NumPolygons-2].IsShadowMask))
-        poly->ClearStencil = poly->IsShadowMask;
-    else
-        poly->ClearStencil = false;
-
     poly->IsShadow = ((CurPolygonAttr & 0x30) == 0x30) && !poly->IsShadowMask;
 
     if (LastStripPolygon && clipstart > 0)
diff --git a/src/GPU3D.h b/src/GPU3D.h
index bebdef0b..6c1b43ce 100644
--- a/src/GPU3D.h
+++ b/src/GPU3D.h
@@ -62,7 +62,6 @@ typedef struct
 
     bool IsShadowMask;
     bool IsShadow;
-    bool ClearStencil;
 
     // data below rather specific to the software renderer
 
diff --git a/src/GPU3D_Soft.cpp b/src/GPU3D_Soft.cpp
index d022e8a4..fe97b5f7 100644
--- a/src/GPU3D_Soft.cpp
+++ b/src/GPU3D_Soft.cpp
@@ -38,6 +38,7 @@ u32 AttrBuffer[256*192];
 // bit30: translucent flag
 
 u8 StencilBuffer[256*2];
+bool PrevIsShadowMask;
 
 // threading
 
@@ -84,6 +85,8 @@ void Reset()
     memset(DepthBuffer, 0, 256*192 * 4);
     memset(AttrBuffer, 0, 256*192 * 4);
 
+    PrevIsShadowMask = false;
+
     // TODO: make it configurable
     if (!RenderThreadRunning)
     {
@@ -803,9 +806,11 @@ void RenderPolygonScanline(RendererPolygon* rp, s32 y)
     else
         fnDepthTest = DepthTest<false>;
 
-    if (polygon->ClearStencil)
+    if (polygon->IsShadowMask && !PrevIsShadowMask)
         memset(&StencilBuffer[256 * (y&0x1)], 0, 256);
 
+    PrevIsShadowMask = polygon->IsShadowMask;
+
     if (polygon->YTop != polygon->YBottom)
     {
         if (y >= polygon->Vertices[rp->NextVL]->FinalPosition[1] && rp->CurVL != polygon->VBottom)