Agricultural systems and problems of sustainability are complex, covering a range of organisational levels and spatial and temporal scales. Integrated assessment (IA) and modelling (IAM) is an attempt to capture complex multi-scale problems. Scale changes and model linking methods (referred to as scaling methods) are important in dealing with these problems but they are often not well understood. The present study aims to analyse scaling methods used in the recently developed multi-scale IA model SEAMLESS-IF which is applied to two case studies of complex agri-environmental problems. The analysis is based on a classification of up- and down-scaling methods which is extended for the purpose of this study. Our analysis shows that scale changes refer to different spatial, temporal and functional scales with changes in extent, resolution, and coverage rate. Accordingly, SEAMLESS-IF uses a number of different scaling methods including data extrapolation, aggregation and disaggregation, sampling, nested simulation and employs descriptive response functions and technical coefficients derived from explanatory models. Despite the satisfactory results obtained from SEAMLESS-IF, a comparative quantitative analysis of alternative scaling methods is still pending and requires further attention. Improved integration of scaling methods may also help to overcome limitations of IA models related to high data demand, complexity of models and scaling methods considered, and the accumulation of uncertainty due to the use of multiple models. In the case studies, the most challenging scaling problem refers to the appropriate consideration of the farm level as intermediate level between the field and market levels. Among the scaling methods analysed, summary models are hardly applied. This is because they are often unavailable due to limited systems understanding and because they may differ depending on the question at stake. The classification of scaling methods used has been helpful to structure this analysis.