The detection of learning styles in adaptive systems provides a way to better assist learners during their training. A popular approach is to fill out a long questionnaire then ask a specialist to analyze the answers and identify learning styles or types accordingly. Since this process is very time-consuming, a number of automatic approaches have been proposed to reduce the number of questions asked. However the length of questionnaire remains an important concern. In this paper, we address this issue by proposing T-PREDICT, a novel dynamic electronic questionnaire for psychological type prediction that further reduces the number of questions. Experimental results show that it can eliminate 81% more questions of the Myers-Briggs Type indicators questionnaire than three state-of-the-art approaches, while predicting learning styles without increasing the error rate.