The prediction of externally visible characteristics (EVCs) is an important investigative tool and a growing field in forensic genetics. The IrisPlex system uses six single nucleotide polymorphisms to predict blue and brown eye colour in humans with over 90% precision. However, the accuracy of this system has not been tested in samples with known genetic admixture. We therefore tested the IrisPlex assay in 64 samples with known and varying levels of Asian-European genetic admixture. Self-declared eye colour information was obtained from participants. The overall accuracy rate for eye colour assignment was 94%, however only 50 samples achieved classification above the 0.7 probability threshold employed. The correct eye colour was predicted in 100% of both blue and brown eye colour samples although none of the green eye colour samples were accurately predicted. When the probability threshold is removed, the accuracy decreases to 84.4% for all 64 samples, with only 94.1% and 85.7% of brown and blue phenotypes predicted correctly. There was also a trend for decreasing accuracy with increasing number of generations since admixture, where individuals with levels of admixture 1:1 and 1:3 were predicted correctly in 96.9% and 87.5% of cases, respectively, and over 1:7 level of admixture correctly predicted in 71.4% of cases. Prediction of EVCs in admixed individuals may present certain difficulties but has been shown to be possible. Larger sample sizes and different types and levels of genetic admixture can improve our knowledge and assist the predictive accuracy in admixed samples.