We develop a Kalman filter for predicting traffic flow at urban arterials based on data obtained from connected vehicles. The proposed algorithm is computationally efficient and offers a real-time prediction since it invokes the connected vehicle data just before the prediction period. Moreover, it can predict the traffic flow for various penetration rates of connected vehicles (the ratio of the number of connected vehicles to the total number of vehicles). At first, the Kalman filter equations are calibrated using data derived from Vissim traffic simulator for different penetration rates, different fluctuating arrival rates of vehicles and various signal settings. Then the filter is evaluated for a variety of traffic scenarios generated in Vissim simulator. We evaluate the performance of the algorithm for different penetration rates under several traffic situations using some statistical measures. Although many of the previous prediction methods depend highly on data from fixed sensors (i.e., loop detectors and video cameras), which are associated with huge installation and maintenance costs, this study provides a low-cost mean for short-term flow prediction only based on the connected vehicle data.