Micro-sensors were used to quantify macro kinematics of classical cross-country skiing techniques and measure cycle rates and cycle lengths during on-snow training. Data were collected from seven national level participants skiing at two submaximal intensities while wearing a micro-sensor unit (MinimaxX™). Algorithms were developed identifying double poling (DP), diagonal striding (DS), kick-double poling (KDP), tucking (Tuck), and turning (Turn). Technique duration (T-time), cycle rates, and cycle counts were compared to video-derived data to assess system accuracy. There was good reliability between micro-sensor and video calculated cycle rates for DP, DS, and KDP, with small mean differences (Mdiff% = -0.2 ± 3.2, -1.5 ± 2.2 and -1.4 ± 6.2) and trivial to small effect sizes (ES = 0.20, 0.30 and 0.13). Very strong correlations were observed for DP, DS, and KDP for T-time (r = 0.87–0.99) and cycle count (r = 0.87–0.99), while mean values were under-reported by the micro-sensor. Incorrect Turn detection was a major factor in technique cycle misclassification. Data presented highlight the potential of automated ski technique classification in cross-country skiing research. With further refinement, this approach will allow many applied questions associated with pacing, fatigue, technique selection and power output during training and competition to be answered.