Fumigation of a passive plume located in or above the entrainment zone (EZ) into a growing convective boundary layer (CBL) has been simulated by large-eddy simulation (LES). Three non-dimensional parameters, alpha(= w(e0)/w(*0)), z(0)/z(i0), and sigma(z0)/z(i0), are used to classify different cases, where w(*0) is the convective velocity scale, w(e0) the initial entrainment velocity, z(i0) the initial CBL height, z(0) the initial plume height, and sigma(z0) is the initial plume half-depth. Forty cases have been run and analysed. The crosswind-integrated concentrations have been compared with existing laboratory data from a saline convection tank. The results show that LES is a promising tool to reproduce fumigation processes. With a relatively coarse grid mesh near the EZ, LES derives reliable results that are in a good agreement with the laboratory data. The first parameter, alpha, containing the effects due to inversion strength, plays an important role in determining C-0(T), the ground-level concentration (GLC) as a function of dimension less time, T. For large alpha (say > 0.03, corresponding to fast entrainment), variation of alpha gives significant change in C-0(T); whereas for a wide range of alpha between 0.01 and 0.02 (corresponding to slow entrainment), C-0(T) is almost independent of alpha. The starting time of fumigation does not vary significantly with the second parameter, z(0)/z(i0) (relative height of plume), although C-0(T) is, in general, smaller for a higher plume. This confirms laboratory findings that the traditional notion of zero fumigation for a high plume (say above 1.10z(i)) is not correct. The effect of the third parameter, sigma(z0)/z(i0), is on the magnitude of C-0(T); thinner initial plumes have higher GLCs. (C) 2002 Elsevier Science Ltd. All rights reserved.