Previous attacks have shown that in-vehicle networks have vulnerabilities and a successful attack could lead to significant financial loss and danger to life. In this paper, we propose a Pearson correlation based anomaly detection algorithm to detect CAN message modification attacks. The algorithm does not need a priori information about the communication: it identifies signals based on statistical properties, finds the important correlation coefficients for the correlating signals, and detects attacks as deviations from a previously learned normal state.