Previous research has demonstrated substantially improved second language (L2) vocabulary learning when spoken word forms are varied using multiple talkers, speaking styles, or speaking rates. In contrast, the present study varied visual representations of referents for target vocabulary. English speakers learned Spanish words in formats of no variability (6 repetitions, 1 image of the word referent), moderate variability (2 repetitions each of 3 different images), and high variability (1 repetition each of 6 different images). In contrast to the positive effects of word-form variability, referent token variability produced negative effects on L2 vocabulary learning. These findings support the type of processing resource allocation model (Barcroft, 2002) because increases in referent-oriented processing diverted available resources from form-oriented processing and learning.