State-and-transition models have received a great deal of attention since the introduction of the concept to range management in 1989. Nonetheless, only recently have sets of state-and-transition models been produced that can be used by agency personnel and private citizens, and there is little guidance available for developing and interpreting models. Based upon our experiences developing models for the state of New Mexico, we address the following questions: 1) how is information assembled to create site-specific models for entire regions, 2) what ecological issues should be considered in model development and classification, and 3) how should models be used? We review the general structure of state-and-transition models, emphasizing the distinction between changes among communities within states (pathways) that are reversible with changes in climate and "facilitating practices" (e.g. grazing management), and changes among states (transitions) that are reversible only with "accelerating practices" such as seeding, shrub control, or the recovery of soil stability and historical hydrologic function. Both pathways and transitions occur, so these models are complementary. Ecological sites and the climatically-defined regions within which they occur (land resource units) serve as a framework for developing and selecting models. We illustrate the importance of clearly delineating ecological sites to produce models and describe how we have dealt with poorly-delineated sites. Producing specific models requires an understanding of the multiple ecological mechanisms underlying transitions. We show how models can represent and distinguish alternative and complementary hypotheses for transitions. Although there may be several mechanisms underlying transitions, they tend to fall within discrete categories based upon a few, fundamental ecological processes and their relation- Dyksterhuis (1949) that is based on the successional theory of Clements (1916) and the edaphic polyclimax concept of Tansley (1935). This model emphasized the return of disturbed communities to a competitively-determined climax state and has been a guiding principle in range management (Westoby 1980). Upon recognizing an undesirable trend in plant community composition, managers could respond by reducing or redistributing grazing pressure and effect a return to desirable conditions. An important reason for the success of this model is that it provided a method to measure and compare land condition against the expectations of the model (i.e., the similarity index), thus providing a concrete link between theoretical expectations and management response.Rangeland managers have long recognized that semiarid grasslands can transform into shrub-dominated states that cannot be returned to grassland through grazing management (Laycock 1991), contrary to applications of the succession-retrogression model. Assuming that a single, competition-defined equilibrium plant community should exist for each site, alternative states, and the rangelands in whic...