As length and timing of the growing season are major factors explaining differences in carbon exchange of ecosystems, we analyzed seasonal patterns of net ecosystem carbon exchange (F_NEE) using eddy covariance data of the FLUXNET data base (http://www-eosdis.ornl.gov/FLUXNET). The study included boreal and temperate, deciduous and coniferous forests, Mediterranean evergreen systems, rainforest, native and managed temperate grasslands, tundra, and C₃ and C₄ crops. Generalization of seasonal patterns are useful for identifying functional vegetation types for global dynamic vegetation models, as well as for global inversion studies, and can help improve phenological modules in SVAT or biogeochemical models. The results of this study have important validation potential for global carbon cycle modeling. The phasing of respiratory and assimilatory capacity differed within forest types: for temperate coniferous forests seasonal uptake and release capacities are in phase, for temperate deciduous and boreal coniferous forests, release was delayed compared to uptake. According to seasonal pattern of maximum nighttime release (evaluated over 15-day periods, F_max) the study sites can be grouped in four classes: (1) boreal and high altitude conifers and grasslands; (2) temperate deciduous and temperate conifers; (3) tundra and crops; (4) evergreen Mediterranean and tropical forests. Similar results are found for maximum daytime uptake (F_min) and the integral net carbon flux, but temperate deciduous forests fall into class 1. For forests, seasonal amplitudes of F_max and F_min increased in the order tropical < Mediterranean and temperate coniferous < temperate deciduous and boreal forests, and the pattern seems relatively stable for these groups. The seasonal amplitudes of F_max and F_min are largest for managed grasslands and crops. Largest observed values of F_min varied between -48 and -2 μmol m^-2 s^-1, decreasing in the order C₄-crops > C₃-crops > temperate deciduous forests > temperate conifers > boreal conifers > tundra ecosystems. Due to data restrictions, our analysis centered mainly on Northern Hemisphere temperate and boreal forest ecosystems. Grasslands, crops, Mediterranean ecosystems, and rainforests are under-represented, as are savanna systems, wooded grassland, shrubland, or year-round measurements in tundra systems. For regional or global estimates of carbon sequestration potentials, future investigations of eddy covariance should expand in these systems.