Monitoring and measuring the behaviour and movement of aquatic animals in the wild is typically challenging, though micro-accelerometer (archival or telemetry) tags now provide the means to remotely identify and quantify behavioural states and rates such as resting, swimming and migrating, and to estimate activity and energy budgets. Most studies use low-frequency (≤32 Hz) accelerometer sampling because of battery and data-archiving constraints. In this study we assessed the effect of sampling frequency (aliasing) on activity detection probability using the great sculpin (Myoxocephalus polyacanthoceaphalus) as a model species. Feeding strikes and escape responses (fast-start activities) and spontaneous movements among seven different great sculpin were triggered, observed and recorded using video records and a tri-axial accelerometer sampling at 100 Hz. We demonstrate that multiple parameters in the time and probability domains can statistically differentiate between activities with high detection (90%) and identification (80%) probabilities. Detection probability for feeding and escape activities decreased by 50% when sampling at <10 Hz. Our analyses illustrate additional problems associated with aliasing and how activity and energy-budget estimates can be compromised and misinterpreted. We recommend that high-frequency (>30 Hz) accelerometer sampling be used in similar laboratory and field studies. If battery and/or data storage is limited, we also recommend archiving the events via an on-board algorithm that determines the highest likelihood and subsequent archiving of the various event classes of interest.