High fidelity models that balance accuracy and computation load are essential for real-time model-based control of homogeneous charge compression ignition (HCCI) engines. Gray-box modeling offers an effective technique to obtain desirable HCCI control models. In this paper, a physical HCCI engine model is combined with two feed-forward artificial neural network models to form a serial architecture gray-box model. The resulting model can predict three major HCCI engine control outputs, including combustion phasing, indicated mean effective pressure (IMEP), and exhaust gas temperature (Texh). The gray-box model is trained and validated with the steady-state and transient experimental data for a large range of HCCI operating conditions. The results indicate that the gray-box model significantly improves the predictions from the physical model. For 234 HCCI conditions tested, the gray-box model predicts combustion phasing, IMEP, and Texh with an average error of less than 1 crank angle degree, 0.2 bar, and 6 °C, respectively. The gray-box model is computationally efficient and it can be used for real-time control application of HCCI engines.