In our study we apply different IRT-models on trend data from 1970 to 2006 in order to analyze the impact of the different models on the results. We use data from the four countries that took part in all five IEA-studies on reading literacy of 9-10 year-old students in the mentioned period. Basically, IRT-models vary in the number of parameters that describe the relation between the latent trait and the observed response behavior on an item. There is an ongoing discussion which model is most appropriate. In recent large-scale assessment studies, for instance, IEA uses the 3-PL model in studies like PIRLS and TIMSS whereas the OECD employs the 1-PL model when analyzing PISA achievement tests. Defenders of the 1-PL model highlight its straightforwardness and simplicity. Those who prefer more complex models argue that the assumptions of the 1-PL model are overly restrictive and do not meet actual data. In our view, in social science models hardly ever fully capture reality (Forster, 2004) and it is hard to weigh the arguments against each other. For pragmatic reasons, we will therefore focus on the question if the chosen IRT-model affects the country level achievement estimates with practical relevance.