There is a belief among research workers that their experiments have two possible outcomes, either they are successful or else the results will have to be taken to a statistician. Although the belief is not universal, it does account for a certain amount of brooding by statisticians on their place in scientific research. This may be why I have made the title I was given, the excuse for examining some of my own activities and attitudes in bringing agricultural research data into juxtaposition with mathematical models. I apologize in advance for the egocentricity.

Statisticians in agricultural research find considerable employment in enabling their customers to adorn papers with a sufficiency of standard errors and significance tests for the satisfaction of editorial boards. Underlying this apparently cosmetic activity is the important function of testing the logic of experimental conclusions and preventing their too easy or too general acceptance. The statistical approach to model building is therefore that of a strength tester. Theoretical assumptions may be vital for the construction of the model but statistical appraisal must be empirical, testing the correspondence between the model and these aspects of experience which it is intended to describe. On the basis of these tests, some parts or some uses of the model may be rejected as unsound and other parts or uses may be accepted, with varying degrees of confidence, as being reliable. Warning notices against undue dependence may be posted on these parts which cannot be tested for lack of data.