Summary Linking psychosocial measures to the cortisol awakening response (CAR) demands accurate saliva sampling times. Monitoring adherence to the saliva sampling protocol requires electronic monitoring of both awakening and sampling times since self-reported times are inaccurate. Delays greater than 15 min between awakening and commencement of saliva sampling reduce CAR magnitude. Less delay has been judged tolerable but remains unexplored for different magnitude measures, and for timing of the CAR peak. Study 1: Fifty healthy females (21 ± 4 years) were instructed to collect saliva on four days at 0, 15, 30 and 45 min post-awakening (samples 1–4). Both self-reported awakening and sampling times were electronically monitored using actigraphy and track caps. Self-reported awakening was later than actigraph estimated awakening (median difference of 4 min). Estimates of CAR magnitude were significantly greater on non-adherent days (delay of 5–15 min) compared to adherent days (delay < 5 min). On non-adherent compared to adherent days cortisol on average peaked earlier, at sample 3 rather than at sample 4. Study 2: Accurately timed cortisol values were obtained in an intensive investigation of 10 participants who collected saliva on 2 days every 5 min for 30 min post-awakening. Cortisol did not significantly increase until 10 min post-awakening, suggesting a time lag may be typical between awakening and observation of a cortisol increase. We conclude that moderate delays between awakening and collection of saliva samples previously considered tolerable result in erroneous estimation of CAR magnitude and timing of the peak. These results are attributed to an approximate 10 min time lag between awakening and the start of the cortisol rise. The absence of this latent period in calculations leads to overestimation of the CAR magnitude on moderately non-adherent sampling days. These findings, if more universally generalizable, will further theoretical understanding of the physiology of the CAR, but are methodologically challenging for researchers since self-reported awakening times are not accurate enough to override the concerns raised. However accurate electronic measurement of adherence to protocol would enable sampling delays to be taken into account in computing CAR estimates.