Fire experiments generally aim to relate fire behavior to fuel and weather conditions. One of the main factors influencing the fire rate of spread is the wind speed, but its measurement in fire experiments is challenging due to the combination of wind-flow turbulence and the remote location of the sensors relative to the fire front. Differences arise because i) the feedbacks of the fire plume on the ambient wind modify the wind speed at fire location, ii) the environment at the measurement location may differ from the fire plot characteristics in non-uniform conditions (fuel, topography), which may induce a spatial bias between mean winds and iii) instantaneous fluctuations at both locations are weakly correlated, which affects the representativeness of wind measurement, because of the relatively brief duration of fire experiments. The present work disentangles the sources of bias and variability in wind measurements (ii and iii), and provides understanding of wind measurement accuracy in fire experiments. The combination of both measurement error sources, both in representativeness and spatial bias, has to date never been studied. In the present study, we employed a modelling approach using HIGRAD/FIRETEC to compute Large-Eddy- Simulations (LES) of wind flows over typical fire experiments. Our simulations show that the spatial bias resulting from the distance to the trailing edge of the measurement location, can be very strong in a realistic configuration and that 7h (with h the canopy height) is too short to get measurements representative of the ambient open wind. In this context, we recommend that the distance to the trailing edge is measured and the upwind canopy characterized (at least in terms of height), so that horizontal Wind Adjustment Factor can be applied to correct the measured wind. Another source of bias is the fact that the fire wind, defined as the average wind at 20ft above the fire plot, can be altered by the presence of the upwind safety break, when its size is larger than 6 to 7h. The representativeness of wind measurements can be estimated by comparing values of fire and measured winds corresponding to replicates of virtual fire experiments. We confirm the finding of two pioneering studies on wind measurement representativeness, which already suggested that measurement errors resulting from both uncertainties in both measured and fire wind could be very significant in typical fire experiments, especially when the plot size was small, when the experiment duration was short, and when the number of sensors was small. For our preliminary results, obtained in more realistic configurations, we found that square fire plots larger than 200 m and a number of five sensors seem to provide a reasonable measurement accuracy, provided that the fire does not spread too fast (which would result in shorter experiment duration) and that the wind is not too low (as the measurement representativeness increases with ambient wind speed)