This paper discusses the main properties of models and model assessments forming the notion of their quality, which include the accuracy, reliability and details of the description of approximate model assessments. On the basis of taking into account the entire set of quantitative parameters of these characteristics, the notion of the measure of model quality is introduced. The main property determining the model quality is highlighted, namely the accuracy of model calculation using independent data. Approximate estimations of the values of a variable in the interval and probabilistic form are considered, which can be constructed on the basis of data on its values in the past. For probabilistic assessments of the values of the variables, it is suggested that, in addition to the measure of accuracy, the measure of informativity should be used, which is determined in terms of entropies of the corresponding probability distribution functions. In developing the algorithms for assessing the measure of quality, special attention is paid to calculating the parameters characterizing the reliability of approximate model assessments and the models themselves. The reliability is assessed in terms of the probabilities of the events occurring when the model is constructed or used. The measure of probability of the model of dependence determined at the stage of constructing it by using the learning sample is calculated as the product of two probabilities: the measure of confidence to the predictors of a dependent variable, and the measure of confidence to the operator describing a dependence of this variable on them. Recommendations are made for assessing these probabilities. N view of the stricter requirements for the reliability of predictive assessments, the algorithm is suggested for increasing it by combining into an ensemble the conditional model and unconditional probability assessments.Keywords: construction of models, verification of hypotheses, properties of approximate assessments, measure of model quality, improvement in predictability.