The search for general properties in the structure of ecological networks is currently a very active area of research. Meta-analyses of published networks are a widely used technique. To have the best chance of discovering common properties though, networks should be constructed using a standardized approach. However, this is rarely the case, and pollination networks are constructed using two main methods: transects and timed observations. To investigate the potential for variation in network structure arising from different construction techniques we constructed plant–pollinator networks using two different methods at a single site, repeating our protocol over three field seasons. Transects and timed observation methods differ in the evenness of observation effort allocated among plant species in the observed community. We show that the uneven allocation of observation effort significantly affects the number of unique interactions in the network, and we reveal a strong trend in effects on web asymmetry and evenness of marginal abundance distributions. However, these effects do not appear to extend to the higher-order properties of connectance and nestedness.