In many bird monitoring surveys, no attempt is made to estimate bird densities or abundance. Instead, counts of one form or another are made, and these are assumed to correlate with bird density. Unless complete counts on sample plots are feasible, this approach can easily lead to false conclusions, because detectability of birds varies by species, habitat, observer and many other factors. Trends in time of counts often reflect trends in detectability, rather than trends in abundance. Conclusions are further compromised when surveys are conducted at unrepresentative sites. We consider how to avoid these problems. We give a brief description of distance sampling methods, which allow detectability to be estimated. We consider strategies to ease their implementation, to enhance their reliability, to adapt the methods for difficult species, and to deal with circumstances in which representative sampling is problematic. We also consider some of the common problems encountered, and suggest solutions.