A method based on dendrochronology to estimate gully erosion rates was developed as an alternative of traditional methods for assessing medium-term gully retreat rates, such as field monitoring of headcuts or aerial-photo interpretation of gully retreat. The method makes use of trees or parts of a tree affected by gully erosion revealing information on the history of the erosion process by datable deviations of their normal growth pattern, hence defined as 'datable objects'. These include roots exposed by erosion; browsing scars made by ungulates on exposed roots or on above-ground parts of fallen trees; exposed and dead root ends; root suckers; stems, branches and leading shoots of fallen trees; and a sequence of trees within a gully. The method is based on the differentiation between three main conditions depending on the relation between the dynamics of the datable object (part of the tree) and the development of the gully. The first condition implies that the datable object was created before erosion of the gully volume to be dated, e.g. exposed tree roots. According to the second condition, the datable object developed as an immediate consequence of the erosion event, e.g. growth reactions of a fallen tree. The third condition implies that the datable object was created some time after the erosion event took place, e.g. trees colonising the gully bed. Each principle has consequences for the accuracy and the correct interpretation of the estimated erosion rate, i.e. whether the true erosion rate is underestimated, exact or overestimated. In spite of methodological limitations and dendrochronological dating problems, the method was successfully applied in southeast Spain. Conservative estimations of gully-head retreat rate resulted in an average medium-term (3-46 years) value of 6 m(3) year(-1) (n = 9). For gully sidewall processes, the average minimum erosion rate per unit sidewall length amounted 0.1 m(3) year(-1) m(-1) (n = 9). A strong correlation was found between the headcut retreat rate (upsilon (m(ortho)) m(3) year(-1)) and the drainage-basin area (A, m(2)) of the gullies, expressed by upsilon (m(ortho)) = 0.02 A(0.57) (R-2 = 0.93, n = 9). Comparing the findings from this study with those obtained by short-term headcut retreat monitoring suggests a high reliability of the estimated retreat rates, supporting the applicability of the developed dendrochronological method. (C) 2001 Elsevier Science B.V. All rights reserved.