This paper aims to create bankruptcy prediction models using logistic regression and neural networks based on the data of Estonian manufacturing firms. The models are composed and tested on the whole population data of bankrupt firms and their vital counterparts for years 2005-2008. Composed models are also tested on the data of firms from economic recession years of 2009-2010. The results indicate that models based on different methods have similar predictive abilities, yet two and three years before bankruptcy they are not as good as for one year before bankruptcy. Also, the models do not perform as well when using data from economic recession years.