Intelligence tests produce composite scores that are interpreted as indexes of psychometric g. Like all measures, general intelligence composites are not pure representations of their intended construct, so it is important to evaluate the score characteristics that affect accuracy in measurement. In this study, we identified three characteristics of general intelligence composite scores that vary across intelligence tests, including the number, the g loadings, and the heterogeneity of contributing subtests. We created 77 composite scores to test the influence of these characteristics in measuring psychometric g. Internal consistency reliability coefficients and g loadings were calculated for the composites. General intelligence composites most accurately index psychometric using numerous highly g-loaded subtests. Considering confidence intervals, composites stemming from four subtests produced scores as highly g loaded as those composites that stem from additional subtests. Discussion focuses on what methods should be use to optimally measure psychometric g and how standards in constructing composites should balance psychometric and practical considerations.