Exceptions to the generality of the stress‐gradient hypothesis (SGH) may be reconciled by considering species‐specific traits and stress tolerance strategies. Studies have tested stress tolerance and competitive ability in mediating interaction outcomes, but few have incorporated this to predict how species interactions shift between competition and facilitation along stress gradients. We used field surveys, salt tolerance and competition experiments to develop a predictive model interspecific interaction shifts across salinity stress gradients. Field survey and greenhouse tolerance tests revealed tradeoffs between stress tolerance and competitive ability. Modeling showed that along salinity gradients, (1) plant interactions shifted from competition to facilitation at high salinities within the physiological limits of salt‐intolerant plants, (2) facilitation collapsed when salinity stress exceeded the physiological tolerance of salt‐intolerant plants, and (3) neighbor removal experiments overestimate interspecific facilitation by including intraspecific effects. A community‐level field experiment, suggested that (1) species interactions are competitive in benign and, facilitative in harsh condition, but fuzzy under medium environmental stress due to niche differences of species and weak stress amelioration, and (2) the SGH works on strong but not weak stress gradients, so SGH confusion arises when it is applied across questionable stress gradients. Our study clarifies how species interactions vary along stress gradients. Moving forward, focusing on SGH applications rather than exceptions on weak or nonexistent gradients would be most productive.