A technique of estimating particulate matter (PM) from gasoline direct injection engines is proposed that is used to compute mass density and particle number density of PM by using fuel mass in rich mixtures obtained by using non-combustion computational fluid dynamics (CFD).The CFD code that was developed by the authors employed a Cartesian coordinates system as a discretization method and large eddy simulation (LES) as a turbulence model. Fuel spray droplets were treated with the discrete droplet model (DDM). The code was verified with some experimental data such as those obtained from in-cylinder gas-flows with a laser Doppler velocimeter (LDV) and in-cylinder fuel concentration with laser induced fluorescence (LIF).PM emissions from a single-cylinder gasoline direct injection engine were measured with an electrical low pressure impactor (ELPI) to determine the model constants that were required in the estimation model. We confirmed that the technique could be applied to various engine operating conditions and fuel spray patterns. Because combustion-phase calculations are not required to estimate PM with the method, it has been a useful way of exploring optimum injection timing and/or spray specifications within short periods.Split injection to reduce PM in a gasoline direct injection engine was investigated with the method. PM was significantly reduced with split-injection by optimizing the injection timing. However, the effect of reducing PM with split injection was greatly affected by the spray patterns. The reason for this was clarified by analyzing the simulation results.