In this paper, a control-oriented soot model was developed for real-time soot prediction and combustion condition optimization in a gasoline Partially Premixed Combustion (PPC) Engine. PPC is a promising combustion concept that achieves high efficiency, low soot and NOx emissions simultaneously. However, soot emissions were found to be significantly increased with high EGR and pilot injection, therefore a predictive soot model is needed for PPC engine control. The sensitivity of soot emissions to injection events and late-cycle heat release was investigated on a multi-cylinder heavy duty gasoline PPC engine, which indicated main impact factors during soot formation and oxidation processes. The Hiroyasu empirical model was modified according to the sensitivity results, which indicated main influences during soot formation and oxidation processes. By introducing additional compensation factors, this model can be used to predict soot emissions under pilot injection. Model parameters were identified using experimental data under a few representative operating points. In order to reduce the cycle-to-cycle variation resulting in the soot estimation noise, a combustion duration calculation method is proposed to estimate CA10 and CA90. This soot model presented in this paper was validated under load-transient operating conditions and agreed with the measurement results with R2 over 0.8 during transient operations, which is sufficient enough for soot prediction as a virtual soot sensor and for control purpose.