We performed Metropolis Monte Carlo and a comprehensive set of reverse Monte Carlo simulations using the same Lennard-Jones liquid to analyze the algorithmic correctness of the latter method. We found that the present reverse Monte Carlo technique is biased because it samples relatively disordered arrangements more often than ordered ones, which leads to a poor acceptance ratio and questionable relevance to the structure of the real system. On the basis of our results, we propose changes in the algorithm. To remove the deficiency we applied a combined acceptance criterion, which in addition to the very restrictive difference minimization procedure, uses an entropy related factor to select from the possible configurations, Further improvement can be achieved if the simulations are performed on the structure factor and the radial distribution function simultaneously, If the radial distribution function is the target function, incorporation of the r-dependent measurement error in the acceptance criterion is essential to obtain results similar in quality to the structure function techniques. The modifications enhance the performance of the reverse Monte Carlo method without increasing the computational effort. (C) 1997 American Institute of Physics.