Form layout understanding is a task of extracting and structuring information from scanned documents, and consists of primarily three tasks: (i) word grouping, (ii) entity labeling and (iii) entity linking. While the three tasks are dependent on each other, current approaches have solved each of these problems independently. In this work, we propose a multi-task learning approach to jointly learn all the three tasks simultaneously. Since the three tasks are related, the idea is to learn a shared embedding that can perform better on all three tasks. Further, the publicly available form understanding datasets are too small, and not ideal to train complex deep learning models. Multi-task learning is an effective method to provide some degree of regularization to the model for such small sized datasets. The proposed model, MTL-FoUn, outperforms existing approaches of learning the individual form understanding tasks on the publicly available data.