The aim of this study was to assess the factorial structure of the Brazilian version of CSAI-2, using several structures suggested in previous studies. Two samples consisted of Brazilian soccer players, 266 from regional level (age 22.87 ± 4.08 years; athletic experience 11.32 ± 4.15 years) and 263 from national level (age 23.16 ± 4.37 years; athletic experience 11.11 ± 4.78 years) were used. The CSAI-2 is a 27-item inventory that measures negativism, physiological activation and self-confidence in a competitive setting. The results of CFA according to the original structure showed some inadequacy of the model. The model proposed by Cox, Martens, Russell (2003), composed of three factors (17-item), demonstrated better adjustment to the regional level sample (χ2/df = 1.871, CFI = .934, GFI = .916, RMSEA = .057), while the model suggested by Coelho, Vasconcelos-Raposo, Fernandes (2007), composed of two factors (18-item), adapted better to the national level sample (χ2/df = 1.701, CFI = .924, GFI = .914, RMSEA = .052). When we analyzed the two samples together, Coelho et al. (2007) was the better model, because it displayed greater invariance. The use of this model was suggested in the assessment of intensity of negative thoughts and the subsequent confirmation of its psychometric properties is recommended.