This study investigated the relationship among intelligibility, comprehensibility, and accentedness in the speech of second language learners of Spanish of varying proficiency in instructed contexts. It conceptually replicated studies by Munro and Derwing (1995a) and Derwing and Munro (1997), who found partial independence among the three speech dimensions but also evidence that proficiency may mediate the relationship between linguistic features of stimuli (e.g., phonemic and grammatical error rates) and speech dimensions. Speech data from 42 second language learners of Spanish recruited from two different universities were elicited via a semispontaneous speaking task: the picture-based narration from the initial study. Amazon Mechanical Turk was used to recruit 80 native Spanish listeners to transcribe and rate extracted utterances. The utterances were coded for grammatical and phonemic errors, goodness of prosody, and speaking rate. Analyses included mixed-effects models that allowed estimation of individual variation across facets of the data, particularly those of listeners.