Proactive interference (PI) may influence the predictive utility of working memory span tasks. Participants in one experiment (N=70) completed Ravens Advanced Progressive Matrices (RAPM) and multiple versions of operation span and probed recall, modified for the type of memoranda (digits or words). Changing memoranda within- or across-trials released PI, but not doing so permitted PI buildup. Scores from PI-build trials, but not PI-release trials, correlated with RAPM and accounted for as much variance in RAPM as unmodified tasks. These results are consistent with controlled attention and inhibition accounts of working memory, and they elucidate a fundamental component of working memory span tasks.