BACKGROUND: Choice egg-count bioassays are a popular tool for analysing oviposition substrate preferences of gravid mosquitoes. This study aimed at improving the design of two-choice experiments for measuring oviposition substrates preferences of the malaria vector Anopheles gambiae senso lato, a mosquito that lays single eggs. METHODS: In order to achieve high egg-laying success of female An. gambiae sensu stricto (s.s.) and Anopheles arabiensis mosquitoes in experiments, four factors were evaluated: (1) the time provided for mating; (2) the impact of cage size, mosquito age and female body size on insemination; (3) the peak oviposition time; and, (4) the host sources of blood meal. Choice bioassays, with one mosquito released in each cage containing two oviposition cups both with the same oviposition substrate (100 ml water), were used to measure and adjust for egg-laying characteristics of the species. Based on these characteristics an improved design for the egg-count bioassay is proposed. RESULTS: High oviposition rates [84%, 95% confidence interval (CI) 77-89%] were achieved when 300 male and 300 blood-fed female An. gambiae s.s. were held together in a cage for 4 days. The chances for oviposition dropped (odds ratio 0.30; 95% CI 0.14-0.66) when human host source of blood meal was substituted with a rabbit but egg numbers per female were not affected. The number of eggs laid by individual mosquitoes was overdispersed (median = 52, eggs, interquartile range 1-214) and the numbers of eggs laid differed widely between replicates, leading to a highly heterogeneous variance between groups and/or rounds of experiments. Moreover, one-third of mosquitoes laid eggs unequally in both cups with similar substrates giving the illusion of choice. Sample size estimations illustrate that it takes 165 individual mosquitoes to power bioassays sufficiently (power = 0.8, p = 0.05) to detect a 15% shift in comparative preferences of two treatments. CONCLUSION: Two-choice egg count bioassays with Anopheles are best done with a two-tier design that (1) implements a parallel series of experiments with mosquitoes given a choice of two identical substrates choices and, (2) uses a single mosquito in each test cage rather than groups of mosquitoes to assess the preference of a test or control solution. This approach, with sufficient replication, lowers the risk detecting pseudopreferences.