Background: Malaria is characterized by frequent mixed infections and extremely high strain diversity, against which hosts do not acquire sterile immunity. The genetically divergent strains circulating at the within-host and population levels bring challenges to efficient disease control. Most molecular epidemiological studies investigate diversity patterns of pathogen isolates from children, usually under the age of five, since children suffer from higher rates of mortality and severe symptoms in endemic regions. However, a higher multiplicity of infection (MOI) in children makes it difficult to resolve the underlying strain genetic structure within hosts and populations, bringing obstacles for studying population genetics of malaria field samples. In this study, we investigate the impact of host age on the inference accuracy of malaria parasite strains and diversity structure, and propose the best age group sampling strategy for molecular epidemiological studies. By integrating an agent-based malaria transmission model with a state-of-the-art computational tool to resolve mixed infections, we compared the SNP-haplotype inference quality from host samples across different age groups and transmission intensities. Inference quality was assessed using four metrics: the recovery of true haplotypes, strain dictionary, shared infections among hosts, and the underlying strain diversity structure from richness estimators. We then validated the inference quality patterns using an empirical dataset from Uganda. Findings: Samples from the old hosts (age>15) show the best inference accuracy compared to young children (age<5) and young children with low levels of mixed infections (age<5, MOI≤3) from multiple aspects. The within-host strain composition is most accurately inferred from old hosts. Regarding strain haplotype dictionary recovery, 91% haplotypes inferred from old hosts are at least with 0.9 identity to the true strains under intermediate transmission intensity, while that percentage drops to 47% in young children. Under high transmission intensity, the percentages of inferred strains with at least 0.8 identity to the true strains are 83% and 31% in old hosts and young children, respectively. Similarly, the sharing network of infections among hosts is most accurately inferred from old hosts. More importantly, strains accurately recovered from old hosts represent frequent strains at the population level. Evaluation of both simulated and empirical datasets indicates that the strain diversity structure inferred from old hosts also yields the best estimates of population-level strain richness. Interpretation: Our results show that malaria parasite isolates from old hosts can provide the most accurate inference of strain haplotypes, shared infections and population-level strain richness. Therefore, a field study with enriched adult sampling in malaria molecular epidemiology helps inform population-level strain composition, infection patterns, and overall diversity, thereby facilitating research on parasite population genetics and improving effective malaria control. Although the traditionally focused young hosts indeed carry more infections, the high levels of within-host mixed infections bring difficulties to strain inference. Instead, old hosts carry fewer strains, but with higher fitness, thus improving the inference accuracy while facilitating practical disease control. Our findings suggest that older hosts may become a new focus for field sampling in future malaria epidemiological research.