Given several related tasks, multi-task learning can improve the performance of each task through sharing parameters or feature representations. In this paper, we apply multi-task learning to a particular case of distance metric learning, in which we have a small amount of labeled data. Consider the effectiveness of semi-supervised learning handling few labeled machine learning problems, we integrate semi-supervised learning with multi-task learning and distance metric learning. One of the defect of multi-task learning is its low training efficiency, as we need all the training examples from all tasks to train a model. We propose an online learning algorithm to overcome this drawback of multi-task learning. Experiments are conducted on one landmark multi-task learning dataset to demonstrate the efficiency and effectiveness of our online semi-supervised multi-task learning algorithm.