Background: Many children with autism spectrum disorders (ASD) struggle to understand familiar words and learn unfamiliar words. We explored the extent to which these problems reflect deficient use of probabilistic gaze in the extra-linguistic context. Method: Thirty children with ASD and 43 with typical development (TD) participated in a spoken word recognition and mapping task. They viewed photographs of a woman behind three objects and simultaneously heard a word. For word recognition, the objects and words were familiar and the woman gazed ahead (neutral), toward the named object (facilitative), or toward an un-named object (contradictory). For word mapping, the objects and words were unfamiliar and only the neutral and facilitative conditions were employed. The children clicked on the named object, registering accuracy and reaction time. Results: Speed of word recognition did not differ between groups but varied with gaze such that responses were fastest in the facilitative condition and slowest in the contradictory condition. Only the ASD group responded slower to low frequency than high-frequency words. Accuracy of word mapping did not differ between groups, but accuracy varied with gaze with higher performance in the facilitative than neutral condition. Both groups scored above single-trial chance levels in the neutral condition by tracking cross-situational information. Only in the ASD group did mapping vary with receptive vocabulary. Conclusions: Under laboratory conditions, children with ASD can monitor gaze and judge its reliability as a cue to word meaning as well as typical peers. The use of cross-situational statistics to support word learning may be problematic for those who have weak language abilities. (Contains 4 figures, 1 table, and 1 note.)