A mixture experiment, is one where the response depends only on the relative proportions of the ingredients present in the mixture. Combines components in various restricted proportions and observes the value of one or more responses for each mixture. These restrictions may result in a small range in terms of the mixture, this can cause difficulties in model fitting arising from ill-conditioning. Most regression packages are now able to fit models that include all of the variables in all but the most highly constrained regions. This could represent an improvement, but it raises some problems about the stability of parameter estimates in the model. As consequence, the information matrix required for computing the parameters of the model can be almost singular. The purpose of this article is to show a comparison between the Kroneker model and the intercept model, in terms of ill-conditioning. We investigate the use of pseudocomponents for improving conditioning in both models. Practical examples are illustrated to support the conclusions.