Many software organizations are turning to employ crowdsourcing to augment their software production. For current practice of crowdsourcing, it is common to see a mass number of tasks posted on software crowdsourcing platforms, with little guidance for task selection. Considering that crowd developers may vary greatly in expertise, inappropriate developer-task matching will harm the quality of the deliverables. It is also not time-efficient for developers to discover their most appropriate tasks from vast open call requests. We propose an approach called PREM, aiming to appropriately match between developers and tasks. PREM automatically learns from the developers’ historical task data. In addition to task preference, PREM considers the competition nature of crowdsourcing by constructing developers’ prestige network. This differs our approach from previous developer recommendation methods that are based on task and/or individual features. Experiments are conducted on 3 TopCoder datasets with 9,191 tasks in total. Our experimental results show that reasonable accuracies are achievable (63%, 46%, 36% for the 3 datasets respectively, when matching 5 developers to each task) and the constructed prestige network can help improve the matching results.