Working memory (WM) refers to ability to store and manipulate information for a period of time. There is evidence of a close relationship between WM and learning. The aim of this study was to investigate the relationship between WM, intelligence quotient (IQ) on school achievement. Individual neuropsychological assessment was conducted following a specific protocol. The study included 227 children aged 7 to 12 years (M = 9.87, SD = 1.34), of which 119 were females, from private and public schools in Salvador, Bahia, Brazil. The data analysis included descriptive and inferential statistics. Academic achievement was evaluated using the Academic Performance Test (TDE), with subtests in reading, writing and arithmetic. The components of WM were evaluated using backward and forward Digit Span tasks and the backward and forward Corsi block-tapping task. The estimated IQ was assessed using tasks vocabularies and block design of the Wechsler scales. The statistical tests used were linear regression analysis (stepwise) and Pearson correlation. The results showed that the best predictors of academic performance were Digit Span Forward, Corsi Block Backward and Digit Span Backward task. The hypothesis of non-multicollinearity was tested, and it was verified that the constructs were independent (VIF < 10 and Tolerance > 0.20). The variance in academic performance explained by WM model was ΔR = 0.36. High and significant correlations were observed between the components of WM and the TDE. The WM score (Digit Span Backward + Corsi Block Backward) and the TDE were correlated (r = 0.54**). These results indicate that WM is a good predictor of academic achievement than IQ and are consistent with other findings showing WM as a predictor of learning or the potential for learning. This result has important implications for education, particularly with respect to intervention.