Data aggregation has been recognized as an effective technique for reducing communication costs while obtaining useful aggregated information. In this paper, we study the crucial problem of delay-constrained data aggregation in vehicular ad hoc networks (VANETs), which has not been well studied in the literature. With the analysis based on real traces, we observe that there is heterogeneity with node contact patterns, which indicates that some nodes contact other nodes more frequently. Motivated by this observation, we propose an approach called aTree. The centralized aTree first constructs a data aggregation tree based on the shortest path tree and then assigns a waiting time budget to each node on the tree based on dynamic programming. We further develop a distributed aTree, in which a shortest path tree is built in a distributed fashion, and nodes determine their waiting time budgets collaboratively. We have performed extensive simulations on real taxi traces, and results show that our aTree schemes incur much lower transmission overhead while achieving the same performance compared with other schemes.