OBJECTIVES: Heterogeneity in meta-analysis is variation of treatment effects between trials that exceeds within-trial variation. Such variation can have many causes, including differences in trial populations, methodology or interventions. Random effects (RE) models are usually fitted to allow for such variation, including a parameter that quantifies such heterogeneity. In the case of Network Meta-analysis (NMA), a common heterogeneity parameter is typically assumed for the network, making its interpretation less intuitive. Our objective is to investigate the impact of dosing differences on the overall heterogeneity of a network. METHODS: A previous NMA analysed in Senn et al. (2011) was re-examined. Ten treatments for diabetes were compared in 26 studies; the outcome measure was change from baseline HbA1c. We have fitted a Bayesian NMA to the network and compared the level of heterogeneity with an extended network, separating doses, resulting in eighteen treatments. Model fit was based on the sum of squared residuals (SSR). We fitted both fixed effects (FE) and RE for the original and extended networks. RESULTS: The SSR is reduced from 148 in the FE model of the original network to 40 in the extended network. For the RE model, the SSR is reduced from 5 in the original network to 3 in the extended network. The original RE estimated the heterogeneity to be 0.26 (95%CrI: 0.14, 0.43). In the extended model, this was reduced to 0.22 (95%CrI: 0.09, 0.44). CONCLUSIONS: Distinguishing between different doses of the same drug in a NMA can reduce overall heterogeneity and improve model fit. However, since heterogeneity can have a multitude of sources, the extended network does not necessarily resolve all heterogeneity. The fact that some variability can be resolved by taking account of dosage does not argue against RE models but underlines that the interpretation of a common heterogeneity parameter is difficult.