The empirical results presented in the literature on the CAPM generally reflect the theoretical flaws of the model in their unconditional form. Thus, two main lines of research have emerged in an attempt to relax some of the model assumptions in its original form, giving rise to so-called multi-factor models and conditional models. In summary, the conditional models are those in which the expected value of the return of an asset is explained conditionally to a set of information available in the previous period, and the sensitivity to the risk factor, beta, as well as the intercept of the equation regression, alpha, are assumed to be time varying parameter. This work has two main objectives: (i) assess comparatively the treatment of conditional CAPM models in state-space form and estimates from the Kalman filter with the residuals of the observation equation in homocedastic and heteroskedastic forms, based on the work of Ortas, Salvador and Moneva (2014); (ii) evaluate how the use of conditional measurements obtained from the conditional CAPM under the approach previously described can improve the current practice of performance evaluation of investment funds from a sample of the Brazilian market. The results obtained indicate that heteroskedastic modeling, from the point of view of the quality of fit for the sample data, is able to provide better results in relation to homocedastic model and corresponding unconditional models, providing better practices for performance evaluation of funds.