A daytime image classifier, which relics on the difference in spectral signatures of the five channels of AVHRR, is proposed. The methodology is based on the preliminary classification of each pixel with the assumption that it is clear and on subsequent tests for cloud detection. The advantage of the preliminary classification is in that the threshold values for spatial variability tests are dependent upon the surface types identified thus allowing more data to be retained over some complex terrains. A fine dynamic cloud filter was designed with the assumption that the absolute value of the near-IR to visible reflectance ratio correlates negatively with surface temperature. The climatic thermal thresholds can then be updated using a clear observation encountered in an array of pixels. Application of the proposed technique is demonstrated on two examples with 1 km resolution AVHRR data and one example with a subset of sampled 30 km resolution AVHRR data from ISCCP. A quantitative validation is done by comparing the cloud amount from ISCCP C1-product with that derived using the present method for the same equal-area grid cells.