Measures of surprise have been recently studied in statistics. This new concept can be used as the first exploratory tool to verify if a model under the null hypothesis fits appropriately. As no alternative models are necessary, the use of the measures of surprise is considered very simple. At the same time, this new alternative to test the goodness of a model cannot replace the full Bayesian analysis. The aim of this project is threshold selection for threshold models. The estimate of the threshold could be investigated by using the measures of surprise. The reason is that no alternative models are specified for non-extreme data, for which the distribution in extreme value theory is unknown, and thus the surprise would be a credible tool.