Scene classification and/or cloud typing of GOES imagery serves many purposes depending on the user's operational or research needs. Two cloud identification algorithms are applied hourly to daytime imagery during a 1-year (approximately) period over an area in the Northeastern Pacific. One method employs pixel classification of cloud types – liquid, supercooled or mixed, glaciated, cirrus, overlapping clouds, and clear – based upon multi-spectral spatial uniformity and contrast tests (explicit physics). The second method applies supervised learning methods in which the spectral and textural characteristics (implicit physics) of historical cloud class (stratus, cirrus, cumulonimbus, etc) samples within hundreds of GOES images are extracted and stored in a database with a 1-nearest neighbor algorithm applied to new samples. While neither method can claim to be ground truth, an improved assessment and confidence of a given pixel's actual cloud type/classification can be achieved based on a comparison of the two method's output. In conjunction with that assessment, the strengths and weaknesses of each method are discovered or confirmed leading, perhaps, to improvements in the algorithms. Finally, a combination product based on the output of both methods may give the best description of the scene in any given GOES image. Statistical comparison results and examples will be presented.