Captive African elephants used in the tourism industry face numerous welfare issues which are often stress related and linked to high numbers of tourists or human-elephant interactions. Elephant welfare is commonly assessed by quantifying faecal glucocorticoid metabolite (fGCM) concentrations, and monitoring stereotypic behaviour, which are either costly or problematic in identifying underlying causes. Self-directed behaviours (SDBs), a form of displacement activity, have been substantially linked with stress and anxiety in primates, and could be a potential but not yet utilised behavioural marker in elephants. We thus explored the usage of several trunk, tail, and body related SDBs in a group of 7 semi-captive African elephants maintained at the Knysna Elephant Park, a tourist destination offering a variety of close contact experiences. Using continuous behavioural sampling (focal following), SDBs were recorded as they occurred, along with the numbers of nearby tourists (<5 m to focal; with 3 levels: None, Low 1–5, and High >5), and the nature of the tourist interaction (5 levels: None, Feed, Touch, Walk, and Ride). We then compared SDB rates with fGCM concentrations (n = 115) determined following observations (1 day and 2 days after behavioural sampling). Data were analysed using Generalized Linear Mixed effects Models. SDB rates (p/min) significantly increased during elephant Walks (p < 0.001) and Rides (p < 0.001), indicating a correlation with their usage during potentially stressful scenarios. The Touch interaction significantly decreased SDBs (p < 0.05), whilst the Feed interaction had a non-significant effect (p > 0.05), which could indicate the element of control in terms of perceived stress is an important component in welfare, as elephants were free to move away during these interactions. Interestingly, SDBs significantly decreased when tourist numbers were High (p = < 0.05), potentially due to greater quantities of high value food. Additionally, no correlation was found between SDBs and fGCM concentrations determined on day one (p > 0.05), and day two (p > 0.05). Hence, SDBs may be correlated more with anxiety or lower-level stress, which is not significant enough to activate GC production, and may therefore act as a coping strategy utilised to maintain physiological homeostasis during anxiety-inducing situations. Additional research would benefit from coupling SDB observations with different forms of physiological assessments to better understand the internal motivations, and formally establish SDBs as a reliable, cost-effective, and non-invasive welfare index to identify stressed individuals in real time and ensure optimum husbandry.