Genotypes of samples from protease inhibitor-naïve patients in Frankfurt's HIV Cohort were analyzed with five tipranavir resistance prediction algorithms. Mean scores were higher in non-B than in B subtypes. The proportion of non-B subtypes increased with increasing scores, except in weighted algorithms. Virtual and in vitro phenotype analyses of samples with increased scores showed no reduced tipranavir susceptibility. Current algorithms appear suboptimal for interpretation of resistance to tipranavir in non-B subtypes; increased scores might reflect algorithm bias rather than "natural resistance."