Under the condition that the size of suspended particles is nonhomogeneous, we studied how filter grain and suspended particles affect filter coefficient in the early stage. A stochastic method was used to study the variation of the initial filter coefficient. Through physical experiment, the collected data include the initial inflow and outflow concentration, the size distributions of particles in suspension and so on. By introducing the standard capture probability function P(d) and the characteristic length of filter bed Lc, this separated the grains’ and particles’ influence on filter coefficient. An example showed that we could use P(d), Lc to ascertain the result of effluent distribution, ratio of Cout to Cin in any depths and equivalent filter coefficient λm. We also studied the filter coefficient λl during the experiment and the first-order derivative of λl.