When using Lagrangian particle dispersion models for modelling turbulent multiphase flows, the standard practice is to use a large, 'statistically significant' sample of particles to determine particle statistics such as concentrations, fluxes, dispersivities or root mean square (RMS) velocities. What is not generally appreciated is that different samples of the same number of (physically identical) particles will produce different results. This means that Lagrangian modellers are experimentalists rather than theoreticians. Although it should be expected that results from large samples are more reliable than that from small samples, no method is in general use at present to quantify this expectation. The approach followed here enables users of such models to determine how reliable the results from such simulations actually are. A strategy is proposed to determine in an efficient way how large the sample size should be to produce results with given confidence limits. The main feature is the need to perform repeated calculations with samples of a given size. Although the proposed strategy provides confidence limits on numerical results, the computational cost should not be greater than current methods and could be significantly less.