ABSTRACT Thigh-worn accelerometry is commonly implemented to measure step cadence. The default activPAL CREA algorithm is a valid measure of cadence during walking, but its validity during running is unknown. The ActiPASS software is designed to analyse tri-axial accelerometry data from various brands. We tested the validity of CREA v1.3 and ActiPASS 2023.12 to measure step cadence against manually-counted steps. Forty-five participants (26♀, 23.4 ± 4.6 years) completed 5 walking (6 min each, 2–4 mph) and 5 running treadmill (5–7 mph) stages (423 total stages completed). Based on equivalence testing, walking cadence (stages 1–5: 92–124 steps/min) from CREA was statistically equivalent (zone: <±2.2% of the manually-counted mean) to manual counts (92–125 steps/min). However, CREA underpredicted cadence during running stages (stages 6–10: 143–135 steps/min) by ~ 11–20 steps/min (p < 0.001). The ActiPASS-derived cadences were equivalent (zone: <±3.3%) to manual counts for all walking stages (99–127 steps/min) except Stage 1 (zone: ±10.5%). ActiPASS underpredicted cadences during running (stages 6–10: 137–153 steps/min) by ~ 10–16 steps/min (p < 0.001) compared to manual counts (stages 6–10: 154–164 steps/min). The CREA v1.3 algorithm is a valid measure of cadence during walking while ActiPASS 2023.12 is a valid measure of cadence during medium-fast walking. Further research is required to improve step cadence estimation across ambulation speeds.