Reverse Monte Carlo (RMC) simulation of liquid water has been performed on the basis of experimental partial pair correlation functions. The resulted configurations were analyzed in various aspects; the hydrogen bond angle distribution, three body correlation and orientational correlation were calculated. The question of the uniqueness of the RMC method was also examined. In order to do this, two conventional computer simulations of liquid water with different potential models were performed, and the resulted pair correlation function sets were fitted by RMC simulations. The resulted configurations were then compared to the original configurations to study how the RMC method can reproduce the original structure. We showed that the configurations produced by the RMC method are not uniquely related to the pair correlation functions even if the interactions in the original system were pairwise additive. Therefore the difference between the original simulated and the RMC configurations can be a measure of the uncertainty of the RMC results on real water. We found that RMC produces less ordered structure than the original one from various aspects. However, the orientational correlations were reproduced rather successfully. The RMC method exaggerates the amount of the close packed patches in the structure, although these patches certainly exist in liquid water.