We use a general ecosystem process model (BIOME-BGC) coupled with remote sensing information to evaluate the sensitivity of boreal forest regional evapotranspiration (ET) and net primary production (NPP) to land cover spatial scale. Simulations were conducted over a 3 year period (1994--1996) at spatial scales ranging from 30 to 50 km within the BOREAS southern modeling subarea. Simulated fluxes were spatially complex ranging from 0.1 to 3.9 Mg C ha-1 yr-1 and from 18 to 29 cm yr-1. Biomass and leaf area index heterogeneity predominantly controlled this complexity, while biophysical differences between deciduous and coniferous vegetation were of secondary importance. Spatial aggregation of land cover characteristics resulted in mean monthly NPP estimation bias from 25 to 48% (0.11--0.20 g C m-2 d-1) and annual estimation errors from 2 to 14% (0.04--0.31 Mg C ha-1 yr-1). Error was reduced at longer time intervals because coarse scale overestimation errors during spring were partially offset by underestimation of fine scale results during summer and winter. ET was relatively insensitive to land cover spatial scale with an average bias of less than 5% (0.04 kg m-2 d-1). Factors responsible for differences in scaling behavior between ET and NPP included compensating errors for ET calculations and boreal forest spatial and temporal NPP complexity. Careful consideration of landscape spatial and temporal heterogeneity is necessary to identify and mitigate potential error sources when using plot scale information to understand regional scale patterns. Remote sensing data integrated within an ecological process model framework provides an efficient mechanism to evaluate scaling behavior, interpret patterns in coarse resolution data, and identify appropriate scales of operation for various processes. ¿ 1999 American Geophysical Union