Our goal is to understand how enzymes adapt to utilize novel substrates. We and others have shown that directed evolution tends to generate enzyme variants with broadened substrate specificity. Broad-specificity enzymes are generally deleterious to living cells, so this observed trend might be an artifact of the most commonly employed high throughput screens. Here, we demonstrate a more natural and effective screening strategy for directed evolution. The gene encoding model enzyme HIV protease was randomly mutated, and the resulting library was expressed in Escherichia coli cells to eliminate cytotoxic broad-specificity variants. The surviving variants were screened for clones with activity against a reporter enzyme. The wild-type human immunodeficiency virus type I protease (HIV PR) is cytotoxic and exhibits no detectable activity in reactions with beta-galactosidase (BGAL). In contrast, the selected variants were nontoxic and exhibited greater activity and specificity against BGAL than did the wild-type HIV PR in reactions with any substrate. A single round of whole gene random mutagenesis and conventional high-throughput screening does not usually effect complete inversions of substrate specificity. This suggests that a combination of positive and purifying selection engenders more rapid adaptation than positive selection alone.