Grand canonical simulations, (specified chemical potential), are performed with a fixed number of particles by coupling variations in the system size to the instantaneous chemical potential determined by virtual test particle methods. This is a modified form of the pseudo-grand canonical method of Mehta and Kofke [Mel. Phys. 86, 139 (1995)]. For dense fluids the method has the advantage of allowing the chemical potential to be specified without inserting particles, and is particularly suited to confined fluids, allowing the geometry to be changed at constant chemical potential. For the test particle determination of the chemical potential a preferential cell sampling method is developed that improves the excluded volume map sampling of Dietrick ct al. [J. Chem. Phys. 90, 2370 (1989)]. It is several orders of magnitude more efficient than crude Monte Carlo sampling, and 4-40 times more efficient than restricted umbrella sampling. The methods are tested on Lennard-Jones fluids, in the bulk at reduced densities up to 0.95, and confined between two planar walls. (C) 1997 American Institute of Physics.