This paper presents data on the variation of G_max with stress level and density for uncemented and cemented sands. This has been obtained using a technique developed by the authors for the continuous monitoring of G_max during consolidation and shearing in triaxial tests. This technique has been used to investigate the variation of G_max for Toyoura sand and a carbonate sand subject to a range of stress paths. The results for these uncemented sands are broadly consistent with other published data but suggest that the effects of mean effective stress, p', are not being correctly predicted by existing empirical relations at higher stress levels. Further tests have been performed on the carbonate sand which has been artifically cemented and prepared at a range of cement contents and densities. It has been found that existing empirical expressions are generally poor at estimating G_max. An equation has been developed that relates G_max to density, stress level and cement content. The paper presents briefly the methodology, some typical results and some comparisons with existing empirical approaches for estimating G_max.