Current measures of ability emotional intelligence (EI)--including the well-known Mayer-Salovey-Caruso Emotional Intelligence Test (MSCEIT)--suffer from several limitations, including low discriminant validity and questionable construct and incremental validity. We show that the MSCEIT is largely predicted by personality dimensions, general intelligence, and demographics having multiple R's with the MSCEIT branches up to .66; for the general EI factor this relation was even stronger (Multiple R = .76). As concerns the factor structure of the MSCEIT, we found support for four first-order factors, which had differential relations with personality, but no support for a higher-order global EI factor. We discuss implications for employing the MSCEIT, including (a) using the single branches scores rather than the total score, (b) always controlling for personality and general intelligence to ensure unbiased parameter estimates in the EI factors, and (c) correcting for measurement error. Failure to account for these methodological aspects may severely compromise predictive validity testing. We also discuss avenues for the improvement of ability-based tests.