Monitoring sex and age ratios (i.e., population composition) is common practice in deer (Odocoileus spp.) management because these estimates are used to determine population vital rates, as well as assess effects of hunting regulations and other management actions. However, there is longstanding recognition of potential biases in sex and age ratios that can arise from differences in deer behavior and sightability, or from methodological issues, such as using roads for sampling. Biologists often use roads for composition surveys because of convenience and roads are often considered to provide coverage of the sampling area and, consequently, unbiased estimates of sex and age ratios. We tested the long‐standing belief that low‐use roads through mule deer (O. hemionus) winter range provided an unbiased sampling scheme by designing and conducting a random‐route survey in conjunction with the road survey through the same winter range in Round Valley, California, USA, during 2015 and 2016. In addition, we conducted a helicopter survey of the same area in 2016. We found adult male:adult female (buck:doe) ratios were 31–45% greater from the random‐route survey compared with the road survey. There was no difference in buck:doe ratios between road and helicopter surveys, but the random‐route survey had 55% greater buck:doe ratios than the helicopter survey. The helicopter survey also had 2.2–4.7 times greater proportions of unclassified bucks than the road or random‐route surveys. There were no differences in fawn:doe ratios among the 3 survey methods. We conducted additional spatial analyses to determine if the buck:doe ratio was greater along random‐routes because bucks were avoiding roads or because the random‐route survey covered different areas. We found that bucks were not avoiding roads; rather, the roads did not adequately cover the winter range area. There were differences between all 3 methods, but our main message is that the assumption that roads provide adequate coverage of winter range and unbiased estimates of population composition should be evaluated. To evaluate that assumption, we recommend conducting ≥1 survey using random routes and comparing composition estimates with road surveys. Further, if information on buck antler‐class is required for management decisions, a ground survey is preferable to a helicopter survey. © 2020 The Wildlife Society.