Wireless sensor networks are widely used in many application domains in recent years. Data collection is a fundamental function provided by wireless sensor networks. How to eciently collect sensing data from all sensor nodes is critical to the performance of sensor networks. In this dissertation, we aim to study the theoretical limits of data collection in a TDMA-based sensor network in terms of possible and achievable maximum capacity. Various communication scenarios are considered in our analysis, such as with a single sink or multiple sinks, randomly-deployed or arbitrarilydeployed sensors, and dierent communication models. For both randomly-deployed and arbitrarily-deployed sensor networks, an ecient collection algorithm has been proposed under protocol interference model and physical interference model respectively. We can prove that its performance is within a constant factor of the optimal for both single sink and regularly-deployed multiple sinks cases. We also study the capacity bounds of data collection under a general graph model, where two nearby nodes may be unable to communicate due to barriers or path fading, and discuss performance implications. In addition, we further discuss the problem of data collection capacity under Gaussian channel model.