Automatic human facial recognition has been an active reasearch topic with various potential applications. In this paper, we propose effective multi-task deep learning frameworks which can jointly learn representations for three tasks: smile detection, emotion recognition and gender classification. In addition, our frameworks can be learned from multiple sources of data with different kinds of task-specific class labels. The extensive experiments show that ourframeworks achieve superior accuracy over recent state-of-the-art methods in all of three tasks on popular benchmarks. We also show that the joint learning helps the tasks with less data considerably benefit from other tasks with richer data.