This study presents a performance comparison between selected white-, gray- and black-box models for indoor temperature prediction in a university building located at the SDU Campus Odense. It was found that the blackboxmodels outperform the gray- and white-box models in most cases, but the accuracy highly depends on the training data in terms of both period and modes of heat transfer covered by the data set. The average mean absolute error for the best performing black-box model was 0.4C as compared to 1.0C and 0.7C for the gray-box and whitebox models, respectively. In terms of accuracy, the graybox models are a reasonable alternative for black-box only in case of short-term predictions, in which their error decreases to around 0.3-0.8C , depending on the room.