This article describes a research method for developing complex applied models which does not sacrifice the high degree of flexibility that such development requires. The method involves a detailed and concrete description of one's intervention model, followed by a carefully documented and measured application. The model is then evaluated and modified in the light of the data, and reapplied. This cycle is repeated many times. The requirements and problems of this method are discussed, particularly in relation to Campbell and Stanley's factors of external and internal validity.