diff --git a/dlls/wined3d/drawprim.c b/dlls/wined3d/drawprim.c
index 02d0144b916..9443877faab 100644
--- a/dlls/wined3d/drawprim.c
+++ b/dlls/wined3d/drawprim.c
@@ -494,12 +494,12 @@ static void drawStridedSlowVs(const struct wined3d_gl_info *gl_info, const struc
 /* GL locking is done by the caller */
 static void drawStridedInstanced(const struct wined3d_gl_info *gl_info, const struct wined3d_state *state,
         const struct wined3d_stream_info *si, UINT numberOfVertices, GLenum glPrimitiveType,
-        const void *idxData, UINT idxSize, UINT startIdx, UINT base_vertex_index)
+        const void *idxData, UINT idxSize, UINT startIdx, UINT base_vertex_index, UINT instance_count)
 {
-    UINT numInstances = 0, i;
     int numInstancedAttribs = 0, j;
     UINT instancedData[sizeof(si->elements) / sizeof(*si->elements) /* 16 */];
     GLenum idxtype = idxSize == 2 ? GL_UNSIGNED_SHORT : GL_UNSIGNED_INT;
+    UINT i;
 
     if (!idxSize)
     {
@@ -513,22 +513,6 @@ static void drawStridedInstanced(const struct wined3d_gl_info *gl_info, const st
         return;
     }
 
-    /* First, figure out how many instances we have to draw */
-    for (i = 0; i < MAX_STREAMS; ++i)
-    {
-        /* Look at the streams and take the first one which matches */
-        if (state->streams[i].buffer
-                && ((state->streams[i].flags & WINED3DSTREAMSOURCE_INSTANCEDATA)
-                || (state->streams[i].flags & WINED3DSTREAMSOURCE_INDEXEDDATA)))
-        {
-            /* Use the specified number of instances from the first matched
-             * stream. A streamFreq of 0 (with INSTANCEDATA or INDEXEDDATA)
-             * is handled as 1. See d3d9/tests/visual.c-> stream_test(). */
-            numInstances = state->streams[i].frequency ? state->streams[i].frequency : 1;
-            break;
-        }
-    }
-
     for (i = 0; i < sizeof(si->elements) / sizeof(*si->elements); ++i)
     {
         if (!(si->use_map & (1 << i))) continue;
@@ -540,8 +524,8 @@ static void drawStridedInstanced(const struct wined3d_gl_info *gl_info, const st
         }
     }
 
-    /* now draw numInstances instances :-) */
-    for(i = 0; i < numInstances; i++) {
+    for (i = 0; i < instance_count; ++i)
+    {
         /* Specify the instanced attributes using immediate mode calls */
         for(j = 0; j < numInstancedAttribs; j++) {
             const BYTE *ptr = si->elements[instancedData[j]].data.addr
@@ -769,11 +753,11 @@ void draw_primitive(struct wined3d_device *device, UINT start_idx, UINT index_co
                         glPrimType, idx_data, idx_size, start_idx);
             }
         }
-        else if (device->instancedDraw)
+        else if (device->instance_count)
         {
             /* Instancing emulation with mixing immediate mode and arrays */
-            drawStridedInstanced(gl_info, state, stream_info, index_count,
-                    glPrimType, idx_data, idx_size, start_idx, base_vertex_index);
+            drawStridedInstanced(gl_info, state, stream_info, index_count, glPrimType,
+                    idx_data, idx_size, start_idx, base_vertex_index, device->instance_count);
         }
         else
         {
diff --git a/dlls/wined3d/state.c b/dlls/wined3d/state.c
index fae79645927..6f0c4f2f50c 100644
--- a/dlls/wined3d/state.c
+++ b/dlls/wined3d/state.c
@@ -4126,7 +4126,7 @@ static void load_numbered_arrays(struct wined3d_context *context,
     int i;
 
     /* Default to no instancing */
-    device->instancedDraw = FALSE;
+    device->instance_count = 0;
 
     for (i = 0; i < MAX_ATTRIBS; i++)
     {
@@ -4146,8 +4146,9 @@ static void load_numbered_arrays(struct wined3d_context *context,
         /* Do not load instance data. It will be specified using glTexCoord by drawprim */
         if (stream->flags & WINED3DSTREAMSOURCE_INSTANCEDATA)
         {
+            if (!device->instance_count)
+                device->instance_count = state->streams[0].frequency ? state->streams[0].frequency : 1;
             if (context->numbered_array_mask & (1 << i)) unload_numbered_array(context, i);
-            device->instancedDraw = TRUE;
             continue;
         }
 
@@ -4290,7 +4291,7 @@ static void load_vertex_data(const struct wined3d_context *context,
     TRACE("Using fast vertex array code\n");
 
     /* This is fixed function pipeline only, and the fixed function pipeline doesn't do instancing */
-    device->instancedDraw = FALSE;
+    device->instance_count = 0;
 
     /* Blend Data ---------------------------------------------- */
     if ((si->use_map & (1 << WINED3D_FFP_BLENDWEIGHT))
diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h
index 3610fc30b51..ddf1fa2e312 100644
--- a/dlls/wined3d/wined3d_private.h
+++ b/dlls/wined3d/wined3d_private.h
@@ -1724,6 +1724,7 @@ struct wined3d_device
     UINT vs_version, gs_version, ps_version;
     DWORD d3d_vshader_constantF, d3d_pshader_constantF; /* Advertised d3d caps, not GL ones */
     DWORD vs_clipping;
+    UINT instance_count;
 
     WORD view_ident : 1;                /* true iff view matrix is identity */
     WORD vertexBlendUsed : 1;           /* To avoid needless setting of the blend matrices */
@@ -1734,9 +1735,8 @@ struct wined3d_device
     WORD inScene : 1;                   /* A flag to check for proper BeginScene / EndScene call pairs */
     WORD softwareVertexProcessing : 1;  /* process vertex shaders using software or hardware */
     WORD useDrawStridedSlow : 1;
-    WORD instancedDraw : 1;
     WORD filter_messages : 1;
-    WORD padding : 5;
+    WORD padding : 6;
 
     BYTE fixed_function_usage_map;      /* MAX_TEXTURES, 8 */