Molecular replacement using search models derived from nuclear magnetic resonance (NMR) spectroscopy has often proved problematic. It has been known for some time that the overall differences in atomic positions (r.m.s.d.) between the crystalline and the solution states of the same protein are of the order of 1-2 A and approach the limit of molecular replacement. In most cases, this structural difference is a result of calculating the NMR structure with insufficient data, yielding an NMR structure of limited accuracy. A systematic case study was performed to investigate the use of NMR models for molecular replacement on the p53 tetramerization domain: NMR search models of varying degrees of accuracy were employed to solve phases for the 1.5 A X-ray diffraction data. An approximate correlation was found between the accuracy of the NMR search model and the clarity and quality of the molecular-replacement solution. It was found that ensemble models perform better than single averaged models and have a larger tolerance in model inaccuracy. Also, distance-derived B factors can improve the performance of single models.