Skills acquisition can be performed using virtual simulators. The issue about assessment of the skills acquired in those environments is a problem not solved since each simulator can collect different interaction data. Data types and distribution can demand specific approaches, i.e, the multiples interaction possibilities between users and virtual simulators do not allow a unified solution. Several assessment models found in the scientific literature are based on Naive Bayes networks, which in turn are based on two kinds of probability measures: the classical one; and that one applied on fuzzy events. In this work, we propose a third one, which use the probability based on fuzzy parameters. The theoretical proposal of a new Naive Bayes network form, which uses Gaussian distribution, is presented and named GAUNB-FP network. It is proposed as the kernel of a new training assessment system and tested considering a bone marrow harvest simulator with three classes of performance provided by experts. Its performance was evaluated, according to decision matrix analysis, Kappa Coefficient and its variance. The results are compared with other four assessment systems based on different networks found in the scientific literature: Naive Bayes Network, Gaussian Naive Bayes Network, Bayesian Network and Multilayer Perceptron Neural Network. The comparison has shown the new assessment system based on GAUNB-FP network provided better results with respect of two of the other networks. It provided also the best results for two of three classes of performance. Therefore, in this comparative experiment, the training assessment system based on GAUNB-FP network presented competitive results.