Heterogeneous multi-attribute group decision-making (HMAGDM) is a complex decision-making problem that widely exists in the real world. However, there is relatively little research on the HMAGDM problems when the attribute set and alternative set are both heterogeneous, and the existing studies still have some limitations, such as the weight calculation is too simple, lacking objectivity and comprehensiveness; the ranking methods does not consider the utility of both group and individuals simultaneously, lacking flxibility and practicality. In order to obtain more effective decision results, a HMAGDM method integrating multi-granulation weighting model and improved VIKOR in uncertain linguistic environment is proposed in this paper. Our contributions can be identified as follows: (1) On the basis of the uncertainty and closeness of uncertain linguistic terms (ULTs), a measure indicator for the effectiveness of experts’ opinions is proposed, and a finest-granulation weight optimization model for experts is established by maximizing the effectiveness; (2) Based on comprehensive consideration of effectiveness and deviation, a bi-objective optimization model is proposed to obtain the multi-granulation weights of attributes; (3) An improved VIKOR method combining the boundedness of ULTs and the multi-granulation weights of attributes is proposed to obtain more stable and effective ranking results. Finally, the case study and comparative analysis illustrate the feasibility and characteristics of the proposed method.