The present study is focused on the behavior of Structural Equation Model (SEM) parameter, standard-error and goodness-of-fit estimates when some or all of the variables on which the model is fitted have been categorized. The behavior of the estimates is studied for different models, underlying and observed distributions, parameter values, sample sizes, number of categorical variables, and SEM programs. The purpose of this study is to determine which are the factors that may disturb the most the estimation of parameter, standard-error or goodness-of-fit values. If the estimation of parameters seems to be satisfactory for polychoric and polyserial procedures, the standard-error estimates are shown to be often biased, especially for models with polyserial correlations. The use of bootstrap is shown here to be a possible solution to this problem.