Although the relationship between sound and meaning in language is assumed to be largely arbitrary, reliable correspondences between sound and meaning in natural language appear to facilitate word learning. Using a set of independently normed pseudoword and shape stimuli, we examined the real-time effects of sound-to-shape correspondences at initial presentation and throughout an extended learning process resulting in high accuracy. In addition to accuracy and response time (RT) measures, we monitored participants' eye movements to investigate the extent to which visual orienting to objects is influenced by the sound symbolic characteristics of novel labels at initial exposure and throughout learning. Over the course of word learning, congruency of sound and shape properties affected both accuracy and RT with higher accuracy and faster responses for congruent than incongruent items. Eye tracking data reveal that congruent targets were fixated faster than incongruent targets throughout learning and that nontargets consistent with the sound symbolic properties of the word remained attractive distracters, even after overt behavioral differences in accuracy disappeared. This demonstrates the sustained influence of sound symbolism and the importance of sensitive, continuous measures of assessing sound symbolic effects in word learning and lexical processing. Arbitrariness resulted in better final individuation performance only when the arbitrary items were more phonologically distinct than the sound symbolic stimuli. These findings suggest that the advantages of sound symbolism may persist beyond early word learning and serve to significantly influence online lexical processing.