Visuospatial perception is thought to be adaptive—ie, hills are perceived as steeper when capacity is low, or threat is high—guiding appropriate interaction with the environment. Pain (bodily threat) may similarly modulate visuospatial perception, with the extent of modulation influenced by threat magnitude (pain intensity, fear) and associated with behaviour (physical activity). We compared visuospatial perception of the environment between 50 people with painful knee osteoarthritis and 50 age-/sex-matched pain-free control participants using 3 virtual reality tasks (uphill steepness estimation, downhill steepness estimation, and a distance-on-hill measure), exploring associations between visuospatial perception, clinical characteristics (pain intensity, state and trait fear), and behaviour (wrist-worn accelerometry) within a larger knee osteoarthritis group (n = 85). People with knee osteoarthritis overestimated uphill (F1,485 = 19.4, P < 0.001) and downhill (F1,480 = 32.3, P < 0.001) steepness more so than pain-free controls, but the groups did not differ for distance-on-hill measures (U = 1273, P = 0.61). There was also a significant group x steepness interaction for the downhill steepness task (F4,480 = 3.11, P = 0.02). Heightened overestimation in people with knee osteoarthritis relative to pain-free controls increased as downhill slopes became steeper. Results were unchanged in a replication analysis using all knee osteoarthritis participants (n = 85), except the downhill steepness interaction was no longer significant. In people with knee osteoarthritis, higher state fear was associated with greater over-estimation of downhill slope steepness (rho = 0.69, P < 0.001), and greater visuospatial overestimation (distance-on-hill) was associated with lower physical activity levels (rho = −0.22, P = 0.045). These findings suggest that chronic pain may shift perception of the environment in line with protection, with overestimation heightened when threat is greater (steeper hills, more fearful), although impact on real-world behaviour is uncertain.