When considering a mixture model, the key question is what should be the components of the model. A mixture model can be created using two different approaches. In the first approach, the model is created manually, based on user-specified components, their weights and parameters of the components. In the second approach, the model is created completely from data. The partitioning will be done automatically, we need not even to specify the number of components.We will compare both approaches addressing the issues of stability and interpretability of the created models. Data concerning the wages of Czech employees collected for more than twenty years will be used in our study.