In this paper we introduce a model evaluation and comparison metric based on the methodology introduced in Faranda et al (2013) to assess biases and their potential origins in a historical model simulation against long-term reanalysis. The metric is constructed by exploiting recent results of dynamical systems theory linking rare recurrences to the classical statistical theories of extreme events for time series. We compute rare recurrences for 100 years daily mean temperatures data obtained in a model with historical greenhouse forcing (the Institut Pierre-Simon Laplace, IPSL-CM5 model) and compare them with the same quantities obtained from two datasets of reanalysis (20 Century Reanalysis and ERA 20C). The period chosen for the comparison is 1900-2000 and the focus is on the European region. We show that with respect to the traditional approaches, the recurrence technique is sensitive to the change in the size of the selection window of extremes due to the conditions imposed by the dynamics. Eventually, we study the regions which show robust biases with respect to all the techniques investigating