Hyperabrupt-junction varactor diodes, for which dopant concentration decreases with distance from the PN junction, possess favorable characteristics in time-varying systems. These diodes can exhibit a more linear relationship between reverse-bias voltage and junction capacitance than their abruptjunction and linear-junction counterparts. This feature makes them attractive for parametrically time-varying circuits when low distortion is desired. However, hyperabrupt-junction diodes are inadequately represented by standard reverse-bias diode models, which were developed primarily for abrupt and linear-junction diodes. We propose a modified diode model that uses a voltagedependent grading coefficient to more accurately reproduce the junction capacitance over a wide range of bias voltage. Using experimentally extracted junction capacitance from ten commercially available varactor diodes, we demonstrate the improved performance of the proposed Polynomial Grading Coefficient (PGC) model compared to the conventional SPICE model with a constant grading coefficient. A sample implementation of the proposed diode model is presented using a Symbolically Defined Device for use within the commercial circuit solver Keysight ADS.