The Government of Canada acquired a national image coverage from the Systeme Pour l'Observation de la Terre (SPOT 4 - 5) satellites that includes four multispectral bands in the visible to shortwave infrared region at 20m spatial resolution. Five years from 2005 - 2010 were necessary to image all of Canada under clear-sky conditions, while acquisition anniversary dates were less important provided the data were imaged during the snow-free period. These data were downloaded from the GeoBase Orthoimage 2005 - 2010 dataset (http://www.geobase.ca/geobase/en/data/imagery/imr/description.html) and used to map 2005 - 2010 land cover south of treeline. Northern Canada has not currently been remapped since circa 2000 due to technical challenges associated with land cover variability and image acquisition dates relative to short summers. This land cover product includes 16 generic classes based on plant functional and a minimum mapping unit of 20m. Radiometric normalization was applied to balance images acquired near mid-summer during the 'peak-of-season' temporal window. The combined Enhancement and Classification by Progressive Generalization methods were used to classify large-area balanced mosaics over twenty mapping zones. Image interpretation was guided using high resolution imagery and other content in Google Earth. Knowledge of land cover spectral signatures, field experience and published reports were also used to assist interpretation in many regions. Remaining images acquired outside the peak-of-season window in early spring or late fall were subsequently classified using decision trees trained on data from overlapping classified peak-of-season images. Accuracy was assessed using ground truth data acquired during several field campaigns conducted with other government departments such as Parks Canada and the Geological Survey. This sample was enhanced using points interpreted in Google Earth as described above to provide a more even spatial coverage of Canada. Overall accuracy assessed at 71% using 1566 reference points, more than half of which were acquired in the field. When assessed using only land cover that was homogeneous within 3 by 3 pixels to account for potential geolocation errors, accuracy increased to 85% for 349 points that were biased towards easily classified classes such as water.