diff --git a/schwab_sort.h b/schwab_sort.h
index 9905e09..a11e0bb 100644
--- a/schwab_sort.h
+++ b/schwab_sort.h
@@ -17,7 +17,12 @@
 /** Below this many elements we do insertion sort */
 #ifndef SCHWAB_INSERTION_THRESHOLD
 #define SCHWAB_INSERTION_THRESHOLD 128
-#endif /* SCHWAB_DELTA_THRESHOLD */
+#endif /* SCHWAB_INSERTION_THRESHOLD */
+
+/** Below this many elements we do insertion sort */
+#ifndef SCHWAB_SELECTION_THRESHOLD
+#define SCHWAB_SELECTION_THRESHOLD 16
+#endif /* SCHWAB_SELECTION_THRESHOLD */
 
 typedef uint32_t sch_rand_state;
 
@@ -65,7 +70,7 @@ inline void sch_insertion_sort(uint32_t *arr, int low, int high) {
 
 /** Simple insertion sort for small cases v2 - not necessarily better */
 inline void sch_insertion_sort2(uint32_t* arr, int low, int high) {
-	for(size_t i = low + 1; i <= high; ++i) {
+	for(int i = low + 1; i <= high; ++i) {
 		uint32_t key = arr[i];
 
 		/* Separate load and compare to expose ILP */
@@ -81,6 +86,23 @@ inline void sch_insertion_sort2(uint32_t* arr, int low, int high) {
 	}
 }
 
+/** Simple SELECTION sort for small cases - not necessarily better */
+inline void sch_selection_sort(uint32_t* arr, int low, int high) {
+	#pragma GCC unroll 2
+	for(int i = low; i < high; ++i) {
+		/* Min-search remaining array */
+		int mini = i;
+		#pragma GCC unroll 4
+		for(int j = i + 1; j < high + 1; ++j) {
+			if(arr[j] < arr[mini]) mini = j;
+		}
+
+		if(mini != i) {
+			schwab_swap(&arr[i], &arr[mini]);
+		}
+	}
+}
+
 /**
  * 3-way partitioning, in middle all the pivot elements.
  *
@@ -316,9 +338,14 @@ static inline void schwab_sort(
 				high = plo - 1;
 			}
 		} else {
-			/* Just do an insertion sort instead */
-			sch_insertion_sort(array, low, high);
-			return;
+			if(high - low > SCHWAB_SELECTION_THRESHOLD) {
+				/* Just do an insertion sort instead */
+				sch_insertion_sort(array, low, high);
+				return;
+			} else {
+				sch_selection_sort(array, low, high);
+				return;
+			}
 		}
 	}
 }