In this research we analyzed the psychometric properties of the Positive and Negative Affect Schedule (PANAS) in samples coming from four independent studies: university students (Study 1, n = 392; Study 2, n = 395), general adult population (Study 3, n = 316), and athletes (Study 4, n = 533). Through confirmatory analyses we evaluated the following models: (a) two-factor model (Watson, Clark, & Tellegen, 1988), (b) three-factor model by Mehrabian (1997), (c) three-factor model by Gaudreau, Sánchez and Blondin (2006) and (d) bi-factor model (Leue & Beauducel, 2011), all of them in their oblique and orthogonal variants. Several models presented an acceptable fit, but only after allowing correlated errors and excluding the items alert and excited. While orthogonal solutions of the three models showed the best fit in Studies 2, 3 and 4, oblique solutions presented the best fit in Study 1. Cronbach's alpha, composite reliability and omega indexes oscillated between .55 and .89. Considering the original theoretical model and the practical utility of the PANAS, we favor the model of two orthogonal factors, excluding the aforementioned items.