diff --git a/source/pixelsearch1x.c b/source/pixelsearch1x.c
index c1c3f377..302eb67f 100644
--- a/source/pixelsearch1x.c
+++ b/source/pixelsearch1x.c
@@ -1,28 +1,49 @@
 #include <emmintrin.h>
 
+// The number of vector registors to utilize in each innermost loop
+#define NR_VEC 4
+
 unsigned int * pixelsearch1x(unsigned int * start, unsigned int * end, unsigned int color) {
+    // The number of elements in each vector register
+    size_t nr_elems_in_vec = sizeof(__m128i) / sizeof(*start);
 
     // Create a vector of four copies of the target color.
     __m128i vcolor = _mm_set1_epi32(color);
 
     // Loop over start pointer with a step of four unsigned integers.
-    while (start < end - 3) {
-
-        // Load four unsigned integers from start into a vector.
-        __m128i vstart = _mm_loadu_si128((__m128i *) start);
-
-        // Compare vstart and vcolor for equality.
-        __m128i vcmp = _mm_cmpeq_epi32(vstart, vcolor);
-
-        // Create a mask from each byte (using the most significant bit) in vcmp.
-        int mask = _mm_movemask_epi8(vcmp);
-
-        // If the mask is nonzero, there is at least one match.
-        if (mask != 0)
-            break;
-
-        // Increment start by four unsigned integers.
-        start += 4;
+    while ( start <  end - (nr_elems_in_vec * NR_VEC) + 1 ) {
+        __m128i vcmp[NR_VEC];
+
+        for ( int i = 0; i < NR_VEC; i++ ) {
+            // Load four unsigned integers from start into a vector.
+            __m128i vstart = _mm_loadu_si128( ((__m128i *) start) + i );
+
+            // Compare vstart and vcolor for equality.
+            vcmp[i] = _mm_cmpeq_epi32(vstart, vcolor);
+        }
+
+        // Use POR because it has a big throughput
+        __m128i vmask = _mm_or_si128(vcmp[0], vcmp[1]);
+        for ( int i = 2; i < NR_VEC; i++ ) {
+            vmask = _mm_or_si128(vmask, vcmp[i]);
+        }
+
+        // If the result is nonzero, there is at least one match.
+        // This combination of MOVMSKBPS + TEST can be replaced with a PTEST when SSE4.1 is available.
+        if ( _mm_movemask_ps((__m128) vmask) ) {
+            int mask = 0;
+            for ( int i = 0; i < NR_VEC; i++ ) {
+                // Create a mask from each dword (using the most significant bit) in vcmp.
+                mask |= _mm_movemask_ps((__m128) vcmp[i]) << nr_elems_in_vec * i;
+            }
+
+            // Return the pointer to the pixel with matching color.
+            // _tzcnt_u32 can be used instead of __builtin_ctz since they're almost identical.
+            return start + __builtin_ctz(mask);
+        }
+
+        // Increment start for the next batch.
+        start += nr_elems_in_vec * NR_VEC;
     }
 
     // Clean up any remaining elements.
@@ -33,4 +54,4 @@ unsigned int * pixelsearch1x(unsigned int * start, unsigned int * end, unsigned
     }
 
     return start; // start == end if no match.
-}
\ No newline at end of file
+}