Objectives: The present study's objective is to understand the causal mechanisms underpinning the recovery of individuals with whiplash-associated disorders (WAD). We applied Bayesian Networks (BN) to answer two study aims: 1) to identify the causal mechanism(s) of recovery underpinning neck-specific exercise, and 2) quantify if the cyclical pathway of the fear avoidance model (FAM) is supported by the present data. Methods: We analysed a prospective cohort dataset of 216 individuals with chronic WAD. Fifteen variables were used to build a BN model: treatment group (neck-specific exercise with or without a behavioural approach, or general physical activity), muscle endurance, range of motion, hand strength, neck proprioception, pain catastrophizing, fear, anxiety, depression, self-efficacy, perceived work ability, disability, pain intensity, sex, and follow-up time. Results: The BN model showed that neck pain reduction rate was greater after neck-specific exercise compared to physical activity prescription (β = 0.59 points/month [P < 0.001]) only in the presence of two mediators: global neck muscle endurance and perceived work ability. We also found the following pathway of variables that constituted the FAM: anxiety, followed by depressive symptoms, fear, catastrophizing, self-efficacy, and consequently pain Conclusion: We uncovered two mediators which explained the mechanisms of effect behind neck-specific exercise, and proposed an alternative FAM pathway. The present study is the first to apply BN modelling to understand the causal mechanisms of recovery in WAD. In doing so, it is anticipated that such analytical methods could increase the precision of treatment in individuals with chronic WAD.