Data are typically discarded when there are inconsistencies in a database, or when information is discrepant or out of range for the biological characteristics of the species being analyzed. However, such loss of information could have considerable implications for genetic evaluation of cattle. Here, we use different models to estimate genetic parameters for age at first calving in order to prevent data elimination. We used five approaches based on trait distribution to define the limits of censure/disposal: linear model, censorship, penalty and missing methods, and threshold model (binary and polychotomous). Data splitting and Pearson correlation were used to evaluate fitting and comparison of models. The lowest heritabilities were estimated for the missing method and the binary model. Exclusion of outliers from the data considerably affects the estimation of genetic parameters and the ranking of sires. Moreover, models that suffered from data elimination generated the worst classifications in terms of the comparison of models. Researchers should be extremely careful when deciding to discard data. For example, an age at first calving up to 72 months could be considered an outlier for other countries or even other areas in Brazil. Although this value is very high, it could be realistic under arid or semi-arid conditions. In these situations, the penalty or censure models are the most appropriate methods of analyses. (C) 2013 Elsevier B.V. All rights reserved.