One neglected possibility is that sometimes, male genitalia are sexual ornaments under pre-copulatory sexual selection driven by female mate choice based on visual assessment. Evidence for female mating preferences based on genital traits is, however, surprisingly rare. In humans, there seems to be stabilizing sexual selection on penis length (e.g. Dixson et al. 2007). Bertin & Fairbairn (2005) demonstrated the existence of pre-copulatory sexual selection for external genital size in water striders, but it is unclear if this is driven by sexual conflict over mating rates or active female choice (Arnqvist & Rowe 2005). Pre-copulatory female choice based on male genitalia is most likely to evolve in taxa with good visual systems and external genitals (e.g. mammals, spiders (pedipalps) and some fishes).

Male standard length (snout to posterior of last vertebrae) was measured (±0.5 mm) under a dissecting microscope to create two size classes: small (S) ≤20 mm and large (L) ≥24 mm representing the upper and lower population quartiles. Gonopodia were then experimentally shortened under anaesthetic. Half of the males in both body size classes had a major reduction treatment (s) and the others had a minor reduction (l) to account for any effects of surgery on female choice or male behaviour (see the electronic supplementary material). Post-manipulation, all males had blunt-ended gonopodia with terminal structures removed that differed only in length (figure 1). We created four distinct male types: Ll, Ls, Sl and Ss. Gonopodia of Ls males were 15 per cent shorter than those of Ll males (mean: 5.65 versus 6.65 mm, s.e. = 0.05) and gonopodia of Ss males were 17 per cent shorter than those of Sl males (mean: 4.52 versus 5.43 mm, s.e. = 0.04). After surgery, males were individually housed in 1 l aquaria. There was no mortality.

We found that, on average, females preferred to associate with larger males, which is consistent with previous studies of G. holbrooki (Bisazza et al. 2001), possibly owing to avoidance of harassment by smaller males (Dadda et al. 2005). However, our results were driven by the strong preference for large males with longer genitalia. Females spent the same amount of time associating with large males with short genitalia as with small males with large or small genitalia. Interestingly, a female preference for longer genitalia was absent when choosing between small males (Experiment 1), even when larger males were absent (Experiment 2). This suggests that the net benefit of producing a longer gonopodium might vary with male body size. One explanation for female choice is that G. holbrooki vary greatly in size at sexual maturation producing considerable overlap between juvenile and adult male body size. Some females might mistake smaller males with shortened genitalia for juveniles and also prefer to associate with a juvenile rather than a small adult male, resulting in no net preference for small males with a longer gonopodium. Another explanation is that female choice is for genetic gains, with correlation selection on insemination ability only favouring longer genitalia in larger bodied males who are less manoeuvrable.

We cannot confirm whether female mate choice generates stabilizing and/or directional selection with respect to the current average male genitalia length, because treatment males only had their genitalia shortened. At present, we lack a technique to experimentally lengthen the gonopodia of free-swimming males. If directional selection for longer genitalia is occurring, as in G. affinis (Langerhans et al. 2005), one might ask why males have not evolved still longer gonopodia? The most likely explanation is a counterbalancing effect of natural selection owing to reduced speed and/or increased predation risk: such trade-offs are well known for male ornaments (Andersson 1994, including poeciliids e.g. Basolo & Alcaraz 2003) and genitalia (House & Simmons 2003). Locomotor costs of non-retractable genitalia occur in G. affinis (Langerhans et al. 2005) and spiders (Ramos et al. 2004).