Syndromic surveillance (SyS) is an important tool for early warning and monitoring of health in human and animal populations, but its use in aquaculture has been limited. Our study objective was to design a SyS system for Atlantic salmon aquaculture and to evaluate its performance in detecting pancreas disease (PD) outbreaks caused by salmonid alphaviruses on farms. We defined SyS outbreak alarms as cases where monthly farm mortality exceeded predefined cutoffs or deviated significantly from expected values based on predictive generalized linear models. These models were trained for each salmon production area in Norway, using data from 2014 to 2017. The outcome variable was fish mortality per farm-month, and input variables were production and environmental predictors, as well as an offset for the number of fish at risk. We also added autoregressive components to explain temporal dependency within fish cohorts. Subsequently, data from 2018 to 2021 was used to parameterize and validate the SyS system’s performance against the current national PD surveillance program, which relies on routine farm-screening tests using molecular techniques and reports of clinical findings. The study covered 19,119 farm-months, involving 1,618 fish cohorts. The performance of our SyS system varied across production areas, with sensitivity ranging from 80.5% to 87.4% and a false alarm rate of 45.3%–53.2%. The absence of alarms was usually observed in farms that were truly negative for PD, i.e., a negative predictive value range of 81.2%–94.0%. The median time for alarms being raised was either in the same month as the current PD surveillance program or 1 month prior or after it. Our results indicate that the SyS system is a valuable tool for monitoring mortality on salmon farms, but alarms are unspecific if evaluated against an individual disease (PD). Increasing the frequency and granularity of mortality reporting might improve the SyS system’s performance.