Preferred running speed is a relevant condition for measuring spatio-temporal parameters and their variability as it theoretically corresponds to an intensity where the energy cost, the biomechanical constraints and the pleasure are optimal. The objective of this study was to evaluate 1) the between-days reliability of spatio-temporal parameters measurements at preferred running speed and 2) the minimal number of steps required to obtain a representative measurement. Thirty-one recreational runners (age:​ 26 ± 5.5 years;​ running experience > 2 years) performed three sessions of treadmill running at preferred running speed separated by 24 h. Spatio-temporal parameters were stride, step, contact and flight times, cadence, step length, duty factor, vertical and leg stiffness. Mean, linear and non-linear measurements of spatio-temporal parameters were computed. Reliability was determined using intraclass coefficient correlation and his 95 % confidence interval lower band (95 %LB). Number of steps required for intra-session representativity were defined as the first number of steps from which all measurements are equal to the measurement at the longer time series length. For all spatio-temporal parameters, mean data showed excellent reliability (95 %LB > 0.90) but most spatio-temporal parameters linear and non-linear variability measurements showed questionable reliability (95 %LB < 0.79). The absolute speed variation at preferred running speed (minimal detectable change 0.8 km/h) may explain these results. Representative values required between 16 and 352 steps for mean value of spatio-temporal parameters, between 16 and 400 steps for linear measurements, and between 336 and 704 steps for non-linear measurements.