Models of community organization involve variations of the top-down (predator control) or bottom-up (nutrient limitation) hypotheses. Verbal models, however, can be interpreted in different ways leading to confusion. Therefore, we predict from first principles the range of possible trophic level interactions, and define mathematically the instantaneous effects of experimental perturbations. Some of these interactions are logically and biologically unfeasible. The remaining set of 27 feasible models is based on an initial assumption, for simplicity, of linear interactions between trophic levels. Many more complex and non-linear models are logically feasible but, for parsimony, simple ones are tested first. We use an experiment in the boreal forest of Canada to test predictions of instantaneous changes to trophic levels and distinguish between competing models. Seven different perturbations systematically removed each trophic level or, for some levels, supplemented them. The predictions resulting from the perturbations were concerned with the direction of change in biomass in the other levels. The direct effects of each perturbation produced strong top-down and bottom-up changes in biomass. At both the vegetation and herbivore levels top-down was stronger than bottom-up despite some compensatory growth stimulated by herbivory. The combination of experiments produced results consistent with two-way (reciprocal) interactions at each level. Indirect effects on one or two levels removed from the perturbation were either very weak or undetectable. Top-down effects were strong when direct but attenuated quickly. Bottom-up effects were less strong but persisted as indirect effects to higher levels. Although the 'pure reciprocal' model best fits our results for the boreal forest system different models may apply to different ecosystems around the world.