The predictions of general-circulation models (GCMs) are sensitive to the assumed cloud overlap within a vertical column of model grid boxes, but until now no reliable observations of the degree of cloud overlap have been available. In this note we derive the overlap characteristics of clouds from 71 days of high vertical resolution 94 GHz cloud radar data in the UK. It is found that, contrary to the assumption made in most models, vertically continuous clouds tend not to be maximally overlapped. Rather, the overlap of clouds at two levels tends to fall rapidly as their vertical separation is increased, and for levels more than 4 km apart, overlap is essentially random. A simple inverse-exponential expression for the degree of overlap as a function of level separation is proposed that could, once results become available from a variety of other locations and seasons, be implemented in current GCMs with relatively little difficulty.