A statistical modeling approach is proposed for the simulation of local paleoclimatic proxy records using general circulation model (GCM) output. A method for model-consistent statistical downscaling to local weather conditions is developed which can be used as input for process-based proxy models in order to investigate to what extent climate variability obtained from proxy data can be represented by a GCM, and whether, for example, the response of glaciers to climatic change can be reproduced. Downscaling is based on a multiple linear forward regression model using daily sets of operational weather station data and large-scale predictors at various pressure levels obtained from reanalyses of the European Center for Medium-Range Weather Forecasts. Composition and relative impact of predictors vary significantly for individual stations within the area of investigation. Owing to a strong dependence on individual synoptic-scale patterns, daily data give the highest performance which can be further increased by developing seasonal-specific relationships. The model is applied to a long integration of a GCM coupled to a mixed layer ocean (ECHAM4/MLO) simulating present-day and preindustrial climate variability. Patterns of variability are realistically simulated compared to observed station data within an area of Norway for the period 1868--1993. ¿ 1999 American Geophysical Union