diff --git a/GPU/Directx9/TransformPipelineDX9.cpp b/GPU/Directx9/TransformPipelineDX9.cpp
index af0b4c9e70..c9721ff655 100644
--- a/GPU/Directx9/TransformPipelineDX9.cpp
+++ b/GPU/Directx9/TransformPipelineDX9.cpp
@@ -494,22 +494,28 @@ void TransformDrawEngineDX9::SoftwareTransformAndDraw(
 				// Scale UV?
 			} else {
 				// We do software T&L for now
-				float out[3], norm[3];
-				float pos[3], nrm[3];
+				float out[3];
+				float pos[3];
 				Vec3f normal(0, 0, 1);
+				Vec3f worldnormal(0, 0, 1);
 				reader.ReadPos(pos);
-				if (reader.hasNormal())
-					reader.ReadNrm(nrm);
 
 				if (!vertTypeIsSkinningEnabled(vertType)) {
 					Vec3ByMatrix43(out, pos, gstate.worldMatrix);
 					if (reader.hasNormal()) {
-						Norm3ByMatrix43(norm, nrm, gstate.worldMatrix);
-						normal = Vec3f(norm).Normalized();
+						reader.ReadNrm(normal.AsArray());
+						if (gstate.areNormalsReversed()) {
+							normal = -normal;
+						}
+						Norm3ByMatrix43(worldnormal.AsArray(), normal.AsArray(), gstate.worldMatrix);
+						worldnormal = worldnormal.Normalized();
 					}
 				} else {
 					float weights[8];
 					reader.ReadWeights(weights);
+					if (reader.hasNormal())
+						reader.ReadNrm(normal.AsArray());
+
 					// Skinning
 					Vec3f psum(0,0,0);
 					Vec3f nsum(0,0,0);
@@ -519,9 +525,9 @@ void TransformDrawEngineDX9::SoftwareTransformAndDraw(
 							Vec3f tpos(out);
 							psum += tpos * weights[i];
 							if (reader.hasNormal()) {
-								Norm3ByMatrix43(norm, nrm, gstate.boneMatrix+i*12);
-								Vec3f tnorm(norm);
-								nsum += tnorm * weights[i];
+								Vec3f norm;
+								Norm3ByMatrix43(norm.AsArray(), normal.AsArray(), gstate.boneMatrix+i*12);
+								nsum += norm * weights[i];
 							}
 						}
 					}
@@ -529,8 +535,12 @@ void TransformDrawEngineDX9::SoftwareTransformAndDraw(
 					// Yes, we really must multiply by the world matrix too.
 					Vec3ByMatrix43(out, psum.AsArray(), gstate.worldMatrix);
 					if (reader.hasNormal()) {
-						Norm3ByMatrix43(norm, nsum.AsArray(), gstate.worldMatrix);
-						normal = Vec3f(norm).Normalized();
+						normal = nsum;
+						if (gstate.areNormalsReversed()) {
+							normal = -normal;
+						}
+						Norm3ByMatrix43(worldnormal.AsArray(), normal.AsArray(), gstate.worldMatrix);
+						worldnormal = worldnormal.Normalized();
 					}
 				}
 
@@ -544,11 +554,12 @@ void TransformDrawEngineDX9::SoftwareTransformAndDraw(
 					unlitColor[2] = gstate.getMaterialAmbientB() / 255.f;
 					unlitColor[3] = gstate.getMaterialAmbientA() / 255.f;
 				}
-				float litColor0[4];
-				float litColor1[4];
-				lighter.Light(litColor0, litColor1, unlitColor, out, normal);
 
 				if (gstate.isLightingEnabled()) {
+					float litColor0[4];
+					float litColor1[4];
+					lighter.Light(litColor0, litColor1, unlitColor, out, worldnormal);
+
 					// Don't ignore gstate.lmode - we should send two colors in that case
 					for (int j = 0; j < 4; j++) {
 						c0[j] = litColor0[j];
@@ -610,20 +621,16 @@ void TransformDrawEngineDX9::SoftwareTransformAndDraw(
 							break;
 						
 						case GE_PROJMAP_NORMALIZED_NORMAL: // Use normalized normal as source
-							if (reader.hasNormal()) {
-								source = Vec3f(norm).Normalized();
-							} else {
+							source = normal.Normalized();
+							if (!reader.hasNormal()) {
 								ERROR_LOG_REPORT(G3D, "Normal projection mapping without normal?");
-								source = Vec3f(0.0f, 0.0f, 1.0f);
 							}
 							break;
 						
 						case GE_PROJMAP_NORMAL: // Use non-normalized normal as source!
-							if (reader.hasNormal()) {
-								source = Vec3f(norm);
-							} else {
+							source = normal;
+							if (!reader.hasNormal()) {
 								ERROR_LOG_REPORT(G3D, "Normal projection mapping without normal?");
-								source = Vec3f(0.0f, 0.0f, 1.0f);
 							}
 							break;
 						}
@@ -639,11 +646,11 @@ void TransformDrawEngineDX9::SoftwareTransformAndDraw(
 				case GE_TEXMAP_ENVIRONMENT_MAP:
 					// Shade mapping - use two light sources to generate U and V.
 					{
-						Vec3f lightpos0 = Vec3f(&lighter.lpos[gstate.getUVLS0()]).Normalized();
-						Vec3f lightpos1 = Vec3f(&lighter.lpos[gstate.getUVLS1()]).Normalized();
+						Vec3f lightpos0 = Vec3f(&lighter.lpos[gstate.getUVLS0() * 3]).Normalized();
+						Vec3f lightpos1 = Vec3f(&lighter.lpos[gstate.getUVLS1() * 3]).Normalized();
 
-						uv[0] = (1.0f + Dot(lightpos0, normal))/2.0f;
-						uv[1] = (1.0f - Dot(lightpos1, normal))/2.0f;
+						uv[0] = (1.0f + Dot(lightpos0, worldnormal))/2.0f;
+						uv[1] = (1.0f - Dot(lightpos1, worldnormal))/2.0f;
 						uv[2] = 1.0f;
 					}
 					break;
diff --git a/GPU/Directx9/VertexShaderGeneratorDX9.cpp b/GPU/Directx9/VertexShaderGeneratorDX9.cpp
index 4041ddde72..7aca9239ff 100644
--- a/GPU/Directx9/VertexShaderGeneratorDX9.cpp
+++ b/GPU/Directx9/VertexShaderGeneratorDX9.cpp
@@ -305,7 +305,7 @@ void GenerateVertexShaderDX9(int prim, char *buffer, bool useHWTransform) {
 			// No skinning, just standard T&L.
 			WRITE(p, "  float3 worldpos = mul(float4(In.position.xyz, 1.0), u_world).xyz;\n");
 			if (hasNormal)
-				WRITE(p, "  float3 worldnormal = normalize(	mul(float4(In.normal, 0.0), u_world).xyz);\n");
+				WRITE(p, "  float3 worldnormal = normalize(	mul(float4(In.normal, 0.0), u_world).xyz);\n", flipNormal ? "-" : "");
 			else
 				WRITE(p, "  float3 worldnormal = float3(0.0, 0.0, 1.0);\n");
 		} else {
@@ -380,7 +380,7 @@ void GenerateVertexShaderDX9(int prim, char *buffer, bool useHWTransform) {
 			WRITE(p, "  float3 worldpos = mul(float4(skinnedpos, 1.0), u_world).xyz;\n");
 
 			if (hasNormal) {
-				WRITE(p, "  float3 skinnednormal = mul(float4(In.normal, 0.0), skinMatrix).xyz %s;\n", factor);
+				WRITE(p, "  float3 skinnednormal = mul(float4(%sIn.normal, 0.0), skinMatrix).xyz %s;\n", flipNormal ? "-" : "", factor);
 				WRITE(p, "  float3 worldnormal = normalize(mul(float4(skinnednormal, 0.0), u_world).xyz);\n");
 			} else {
 				WRITE(p, "  float3 worldnormal = mul( mul( float4(0.0, 0.0, 1.0, 0.0), skinMatrix), u_world).xyz;\n");