The use of extensive questionnaires has the limitation of filling time bias, related to the ability to focus and accurately respond to many items, justifying the necessity for a brief version. This study aimed to build a brief version of the Caffeine Expectancy Questionnaire in Brazil (B-CaffEQ-BR) composed of 21 items divided into seven factors, with as adequate consistency and reproducibility as the full version. Quantitative procedures using statistical modeling were applied using the CaffEQ-BR (full version) database keeping the Mean Absolute Error (MAE) (based on the full version) <0.5 and Cronbach's α and Intraclass Correlation Coefficient (ICC) ≥0.7. The expert panel (n = 3), in a blind design, evaluated the semantic structuring within the options indicated by previous statistical modeling until the agreement of the expert panel. The participants (n = 62), Brazilian adults who were regular caffeine consumers (175.8 ± 94.4 mg/day), of whom 62.9% were women, 33.1 ± 9.7 years, 24.5 ± 3.8 kg/m2, and 62.9% of whom self-identified as white, were asked to respond twice to the online questionnaire in 48–72 h. The first sample (n = 40) tested interobserver reproducibility with the double application of B-CaffEQ-BR. Another sample (n = 22) answered the CaffEQ-BR (full version) and B-CaffEQ-BR, and the last sample (n = 18) performed the reverse process. The B-CaffEQ-BR presented excellent internal consistency (Cronbach's α ≥ 0.729) and overall reproducibility (ICC ≥ 0.915) for the entire questionnaire and its seven factors. The B-CaffEQ-BR can be a valuable tool in caffeine research with the Brazilian adult population.