1. CONTEXT For arable crops, weeds are the most harmful pests among those targeted by pesticides. They are also crucial sources of plant biodiversity in agricultural landscapes, and host and feed wild fauna. As there is, to date, no curative weed control method as effective as herbicides, reducing herbicide use while limiting the damage to crop production requires combining many partially-efficient cultural practices at the cropping-system scale. 2. OBJECTIVE The objectives were to (1) compare the performances and applicability of cropping systems obtained with three contrasting design approaches, aiming at reducing weed harmfulness for crop production and herbicide use, (2) identify the technical determinants of a more sustainable weed management. The design approaches were: (1) de novo design by diverse experts working in five regions, (2) step-by-step improvement of a series of farmers' systems from a single production context over several years; (3) redesign of one current system by one group of farmers with weed issues, helped with a board game (Mission Ecophyt’eau®) and a decision-support system to evaluate weed impacts (DECIFLORSYS). 3. METHODS We assessed (1) weed impacts on crop production and biodiversity by simulating current and innovative cropping systems with the FLORSYS model over 30 years and with 10 weather scenarios, (2) sustainability by scoring the economic, social and environmental performances of all systems with the multicriteria evaluation tool DEXiPM. To account for production context, we assessed the differences in indicator values of innovations vs the local current system ("trajectories"). 4. RESULTS AND CONCLUSIONS The trajectories showed no trade-off between weed harmfulness for crop production and herbicide use intensity. Reduced herbicide use (-1.0 Treatment Frequency Index) was compensated by combining alternative levers targeting weed prevention, particularly crop diversification (+ 3 to 5 new crops on average) and tillage. Changes in performances were greater with expert design than with the two farmer-design approaches. However, social sustainability from farmers' perspective mostly decreased with expert design (6 trajectories out of 9). Moreover, the best trajectories for reconciling low weed harmfulness and low herbicide use with higher sustainability were obtained by farmers after 10 years of step-by-step design (2 trajectories out of 30). 5. SIGNIFICANCE The use of models such as FLORSYS, DEXiPM, board games and decision-support systems in the two farmers' groups showed that bringing new knowledge in different ways is essential to actually change cropping systems. Implicating farmers early in the design process also improved their acceptance of innovations.