The purpose of this experiment was to determine whether working memory (WM) span task performance is driven by predominantly domain-general or domain-specific processes. Fifty-two participants were administered six WM span measures modified from Kane et al. (2004) to include a suppression component. Three tasks involved spatial recall elements, and three involved verbal elements. Confirmatory factor analyses were conducted to compare the fit of two competing models: a one-factor domain-general model characterized by a unitary WM construct, and a two-factor domain-specific model characterized by separate spatial and verbal WM constructs. Results revealed that the domain-specific model provided a good fit for the data, whereas the domain-general model did not, and the two-factor model provided a significantly better fit than the unitary model. Furthermore, verbal and spatial WM capacity was only moderately correlated in the two-factor model sharing 24-25% variance. The findings that predominantly domain-specific processes drove performance on the WM tasks suggest a reformulation of the current WM framework.