The estimated prevalence of a malaria parasite, Plasmodium mexicanum, of western fence lizards, Sceloporus occidentalis, was compared using two techniques: microscopic examination of blood smears, and nested PCR amplification of the 18S small subunit rRNA gene. Two sites in northern California, USA were investigated, one with known long-term high prevalence of the parasite (30% by blood smear scanning), and one with low prevalence (6%). The nested PCR readily detected very low-level infections (< 1 parasite per 10 000 erythrocytes); such infections are often subpatent by normal microscopic examination. False negatives (scored as not infected after scanning the blood smear, but found infected via PCR) were rare at both sites (4% at the high-prevalence site, 6% at the low-prevalence site). However, a greater proportion of infections was detected only by PCR at the low-prevalence site (50% vs. 9%). If 50% of the infections sustain very weak parasitaemia where lizards are rarely infected, this would accord with hypotheses that predict that parasites should reduce infection growth when transmission is uncommon. The study demonstrates that PCR is a powerful tool to detect very low-level malarial infections in vertebrate hosts, including those with nucleated erythrocytes.