In this paper, a novel region-based approach for estimating the speed of a vehicle using wireless sensor networks is presented. Compared with a point-based approach, which is used to determine the vehicle arrival and departure points separately in each node, the proposed region-based approach is used to determine the speed of a vehicle at the server using the similarities among the sensor data received from two sensor nodes. In the proposed approach, a moving-average filter is applied to reduce noise in the sensor reading. Next, N-samples of data around a feature point with a first-order derivative larger than the chosen threshold are recorded. Delta coding is then applied to compress the data and minimize the power consumption required for communication from the sensor nodes to the server. Finally, the similarity of the data received from the two sensor nodes is measured to estimate the speed of the vehicle. More specifically, a similarity measure, a modified version of a cross-correlation function, is proposed. In addition, an evolutionary programming technique is adopted to find the optimal parameters for the threshold value of the first-order derivative and the number of samples that need to be sent to the server. Experimental results are provided to show the effectiveness of the proposed region-based vehicle speed estimation approach.