ABSTRACT. All indications point to a rapid decline of forest elephants in Central Africa. But traditional forest elephant monitoring methods do not allow accurate estimation of the rate of decline, precise attribution of causes of the decline, or efficient assessment of management or policy actions designed to stem the decline. This paper describes a new, more powerful statistical framework for monitoring forest elephant trends. The framework combines a Markov model of temporal trends witha likelihood-based description of spatial trends in elephant distribution. Using small survey segments, rather than total population size, as the unit of analysis allows the modeling of covariates that strongly influence density, but vary on a relatively small spatial scale. Modeling of spatial gradients in elephant distribution not only improves the power of temporal trend tests, but also provides muchmore insight into the causes of temporal trends than does an analysis of abundance alone. Likelihood-based model selection procedures are advocated as a superior alternative to the classic hypothesis testing format. Data from an ongoing survey of the Gamba Complex of protected areas in Gabon are used to demonstrate how model selection procedures based on Akaike's information criterion can be implemented. Non-parametric analysis of spatial autocorrelation is suggested as a means for detecting trends that are not associated with known covariates and for the identification of previously unknown covariates. Suggestions on other priorities for implementing forest elephant monitoring are discussed.