Polymorphisms of N-acetyltransferase 2 (NAT2) are well known to modify urinary bladder cancer risk as well as efficacy and toxicity of pharmaceuticals via reduction in the enzyme's acetylation capacity. Nevertheless, the discussion about optimal NAT2 phenotype prediction, particularly differentiation between different degrees of slow acetylation, is still controversial. Therefore, we investigated the impact of single nucleotide polymorphisms and their haplotypes on slow acetylation in vivo and on bladder cancer risk. For this purpose, we used a study cohort of 1,712 bladder cancer cases and 2,020 controls genotyped for NAT2 by RFLP-PCR and for the tagSNP rs1495741 by TaqMan(®) assay. A subgroup of 344 individuals was phenotyped by the caffeine test in vivo. We identified an 'ultra-slow' acetylator phenotype based on combined *6A/*6A, *6A/*7B and *7B/*7B genotypes containing the homozygous minor alleles of C282T (rs1041983, *6A, *7B) and G590A (rs1799930, *6A). 'Ultra-slow' acetylators have significantly about 32 and 46 % lower activities of caffeine metabolism compared with other slow acetylators and with the *5B/*5B genotypes, respectively (P < 0.01, both). The 'ultra-slow' genotype showed an association with bladder cancer risk in the univariate analysis (OR = 1.31, P = 0.012) and a trend adjusted for age, gender and smoking habits (OR = 1.22, P = 0.082). In contrast, slow acetylators in general were not associated with bladder cancer risk, neither in the univariate (OR = 1.02, P = 0.78) nor in the adjusted (OR = 0.98, P = 0.77) analysis. In conclusion, this study suggests that NAT2 phenotype prediction should be refined by consideration of an 'ultra-slow' acetylation genotype.