Recommender　systems　are　used　to　recommend　items　for　user　in　e-commerce　with　information　overload.　Utility-based　recommender　systems　build　multi-attribute　utility　function　of　user　and　recommend　the　highest　utility　item　for　user.　Some　utility-based　recommender　systems　use　rating　for　items　to　extract　utility　function,　which　produce　significant　burden　for　user.　The　paper　proposes　a　utility-based　recommender　technique　which　can　predict　attribute　value　utility　and　implicit　holistic　utility　rate　of　items　by　user　browsing　behavior　and　genetic　algorithm,　and　elicit　the　attribute　weight　by　genetic　algorithm,　and　building　a　multi-attribute　utility　function.　The　experimental　results　on　clothing　recommendation　show　that　the　proposed　method　is　superior　to　current　utility-based　methods　on　accuracy,　satisfaction,　usefulness　and　time　expense.