Objective: To determine whether WAIS-IV intelligence scales are predicted by KeyMath performance. Method: Data from 29 males and 53 females was collected from a de-identified neuropsychology database in a South Florida community clinic representing psychological and neurological disorders and non-psychiatric controls. Participants averaged 27.88 years (SD = 10.98), with an average education of 13.56 years (SD = 1.75); 46% Caucasian. Separate, backwards-stepwise linear regression analyses were performed using all KeyMath (version 1) subtests as predictors of WAIS-IV scales FSIQ, PIQ, WMIQ, VCIQ, and PSIQ with a p < .001 significance level. Results: Results of the regression analysis model predicting FSIQ was significant, (F = 9.941, R2 = .620, p < .001). Geometry (t = 3.48) and Measurement (t = 3.51) subtests significantly predicted the model. The model predicting PIQ was significant, (F = 7.82, R2 = .461, p < .001) without significant model predictors. The model predicting WMIQ was significant, (F = 9.76, R2 = .744, p = .001) with Numeration (t = 3.48) and Basic Concepts (t = −3.36) significantly predicting the model. The model predicting VCIQ was significant (F = 6.29, R2 = .343, p < .001) with the Geometry subtest (t = 3.42) a significant model predictor. The model predicting PSIQ was not significant. Conclusion: Verbal and Full-Scale intelligence were best predicted by math tests associated with combined verbal and non-verbal intelligence. High performance on Geometry and Measurement requires visuo-spatial manipulation and unit comparison more than in more verbal subtests. The ability to accurately perform rote arithmetic operations appears to reflect narrow strengths in auditory attention only. Visuo-spatial math ability reflects integration of other necessary cognitive processes, including spatial manipulation and logico-grammatical comprehension. KeyMath is a completely untimed test, and therefore did not predict processing speed.