Purpose: The research is related to solvency forecasting methods and their models, their possible application and determination of precision. The objectives of the study are to examine the ten most commonly used models of insolvency, their application in Latvian companies, and to enable manufacturing companies to continue their business activities with a certain degree of reliability. It is necessary to predict probable insolvency in a timely manner and evaluate it in order to comply with the principle of continuation of business. Design/methodology/approach: The study is conducted on the basis of Latvian manufacturing companies with an annual turnover of 1 million EUR to 3 million EUR over the period of 2011 to 2016. The research methods are as follows: monographic, graphical, analysis of statistical data, correlation and comparative analysis. Findings: According to the results, the lowest Type I error of only 7% was shown by the Skiltere and Zuka model and the Type II error for the model was 29%. The highest classification abilities were shown by the while the Zmijewski, Altman Z'' and Lis models. Type I errors were from 11% to 22%, and the Type II errors were higher, from 33% to 55%. The authors of this paper suggest using other cut-off points than the model proposed by the above authors, thus significantly increasing model accuracy. Research limitations/implications: The paper examines discriminant analysis-based models and the validation is performed by estimating two types of errors and using ROC curves. A Type I error occurs when a model does not predict bankruptcy. A Type II error means that the model has predicted a solvent company as bankrupt. Practical implications: The results of the study comprise the authors considerations of the models as recommended, possible and not recommended for predicting insolvency of Latvian manufacturing enterprises. Originality/value: The authors of the paper not only focus on the results of validation using a two-error method but also analyse the classification abilities of insolvency prediction models determined by area under the ROC curve. The study is also innovative because the authors suggest changing the cut-off points for some models to increase the accuracy of the prediction.