Efficient data delivery in vehicular networks has received increasing attention in recent years. Existing routing protocols for vehicular networks can be loosely divided into two classes: road based routing (RBR) and road oblivious routing (ROR). RBR finds a routing path along roads while ROR does not explicitly forward packets along roads. We have the observation that using either of an RBR algorithm or an ROR algorithm alone in a realistic vehicular network setting leads to deficiency. This results from the fact that network conditions can be different at different locations and evolving over time. Motivated by this observation, this paper proposes an adaptive routing algorithm called RWR that adapts its routing strategy to network dynamics as the packet travels from the source to the destination. Extensive simulations based on a large dataset of real vehicular traces collected from around 2,600 taxis in Shanghai have been conducted. Comparison study shows that RWR produces higher delivery ratio than TSF and GPCR, representative routing algorithms of RBR and ROR, respectively. It achieves low delivery delay at the same time.