Due to their impacts on biodiversity and human health, reducing pesticide use is a key step for transitioning to sustainable farming systems (Matson et al, 1997; Storkey et al, 2012). Following the European SUD Directive, the Ecophyto plan set the objective of halving pesticide use in France, primarily by 2018, but thereafter postponed to 2025 (ECOPHYTO II+ Plan, Ministry of Ecological and Solidarity Transition, 2018). The Ecophyto plan launched the DEPHY network, based on about 3.000 farms engaged in the demonstration that reducing pesticide use is possible without impairing farm profitability. Farms are coordinated in groups of 10-15 farms, and detailed practices data are collected and gathered into a national database (DEPHY FARM network, 2018). Previous research based on this database has shown that in a majority of sites, low pesticide use was not conflicting with profitability nor productivity (Lechenet et al, 2017). Cropping systems with low pesticide use were associated with combinations of preventive measures that varied across production situations and across farms (Lechenet et al, 2016). These results, based on a synchronic comparison of farms, tend to demonstrate that using little amount of pesticide is possible in most cases, but research remains to address the transition from high to low pesticide use, based on a diachronic approach. With a large range of data (more than 1 300 farms having joined the network since at least 3 years), we are performing detailed analysis of the evolution of pesticide use over time, and its relation to the evolution of farm practices implementation. We consider how the production context (soil type, climate, access to specific markets) influences both practices and pesticide use. For each site, we computed a range of variables describing (i) the production situation, (ii) the cropping system and pesticide use at the enrolment in the network, and (iii) the changes in cropping system and pesticide use after a few years. The Treatment Frequency Index (a commonly used index for estimating the pesticide use dependency of a farm (Brunet et al, 2008)) is used as a metric of pesticide use. Variables describing cropping systems include descriptors of practices expected to have an effect on pest pressure, and therefore on pesticide use (e.g. among others: soil tillage strategy, crop sequence, sowing dates, Fertilization level, mechanical weeding, etc. (Davis et al, 2012; Lechenet et al, 2016). A CART (Classification And Regression Tree) method is used to analyse changes in pesticide use (DTFI = TFI final – TFIinitial) as a function of all variables describing (i) changes in cropping systems and crop management, (ii) initial cropping systems and (iii) production situations. The methods allow to identify which evolution of (combinations of) practices are associated with325 a reduction or an increase in pesticide use, taking into account both specificities of production situations and initial cropping systems. Analyses are currently in progress. At the date of the conference, we will present the results. We anticipate that major changes in practices allowing to reduce pesticide use will be in accordance with strategies with low reliance on pesticides identified by Lechenet et al. (2016), namely the diversity of the crop sequence, the soil tillage strategy, the moderation of fertilization, the use of resistant cultivars, the sowing date, the reduction of pesticide dose at each application, etc. We intend to complement this study by analyzing the consequences of these changes in cropping system and management on other sustainability indicators (productivity, profitability at the farm level, work load, environmental impacts)