The analysis of phenomena related to cognitive processes is a very important trend in Natural Language Processing (NLP) research. The study of sentiments, emotions or humour, through language, are a sample about how this tendency acquires a greater importance in the area. In this paper, we present the results obtained on a study of Automatic Humour Recognition (AHR) performed on a corpus of children's texts. The results indicate that through the identification of semantic and affective features the text classification can be achieved with success.