Drug resistance in Mycobacterium tuberculosis is often linked to specific mutations in a limited number of resistance genes. Detection of these mutations in a cultured isolate can predict the resistant phenotype. Genotypic analysis of the mycobacteria directly in a clinical specimen would result in considerable time saving for resistance prediction.To find out whether resistance-predicting genotypes of mycobacteria found after cultivation always give a good reflection of those in the original clinical sample.Restriction fragment length polymorphisms of repetitive polymerase chain reaction (PCR) amplification and cloning of PCR products were used as nonintegrative methods to describe the composition of katG, rpsL and embB genotypes involved in resistance to isoniazid, streptomycin and ethambutol, respectively, in the original sample. This result was then compared to the phenotypic resistance profile after cultivation.Using both methods, mixed, heteroresistant populations could be detected in almost every fifth analyzed sample (katG: 5 of 16; rpsL: 3 of 17; embB: 1 of 21). Direct sequencing, a widely used integrative method, repeatedly failed to detect heteroresistance.Heteroresistance is a valid phenomenon in clinical tuberculosis. It is not rare and not restricted to a particular resistance gene, and is obscured by cultivation as well as by some, not all, culture-independent resistance prediction tests.