OBJECTIVES In meta-analysis, model selection is an important criterion which needs to be tested and validated by strong statistical evidence. The Cohran’s Q-test allows in theory to decide between random-effect and fixed-effect models. Due to the highly conservative nature of this test, three statistics have been built to estimate the heterogeneity between studies to lead the model decision: the I, the Hand the RMETHODS Based on the global formulation of the Cochran’s Q-test, we proposed to analyse jointly the first error species and the second error species in different scenarios based on the number of studies included in each meta-analysis. The goal was to determine the reliability of the Q-test in extreme situations but also to give some benchmark for the reliability of this test. We use simulation methods to analyse the three different methods for calculating the between-study variance compared to the real value of heterogeneity. We also compared different arbitrary levels for model selection using these statistics in different scenarios. RESULTS The Cochran’s Q-test is too conservative with a large number of studies and concludes to the presence of heterogeneity whatever the situation is when the number of studies is higher than 18. In comparison, the different statistics have an average value conversely linked with the number of studies in case of non-heterogeneity: the higher the number of studies, the lower the statistics’ average values. CONCLUSIONS The I and Hstatistics can eventually enhance the use of Cochran’s Q-test by solving conservative issue associated with this test. The model selection can, eventually, be led by benchmark of these statistics jointly with the Cochran’s Q-test.