Abstract
We have tested three methods of estimating the level of a coming season's mean temperature at a station where the statistical association between two selected seasons is as high as one can expect in extratropical regions. The methods are contingency tables, regression equations, and the use of the last few decades if there is a trend at the station which will separate the mean of these decades a fair distance from the long-term mean. A moderate amount of skill was achieved, but the degree of seasonal association in our test case was exceptionally high, and generally these methods will provide only a small improvement over a probability based on knowing only the observed frequency distribution.