The aim of the present study was to re-examine the taxonomic position and structure of taxon K (also known as group K) within the Burkholderia cepacia complex (Bcc). For this purpose, a representative set of strains was examined by a traditional polyphasic taxonomic approach, by multilocus sequence typing (MLST) analysis and by analysis of available whole-genome sequences. Analysis of the recA gene sequence revealed three different lineages, designated recA-I, recA-II and recA-III. DNA–DNA hybridization experiments demonstrated that recA-I and recA-II isolates each represented a single novel species. However, DNA–DNA hybridization values of recA-II strains towards recA-III strains and among recA-III strains were at the threshold level for species delineation. By MLST, recA-I isolates were clearly distinguished from the others and represented a distinct lineage referred to as MLST-I, whereas recA-II and recA-III isolates formed a second MLST lineage referred to as MLST-II. A divergence value of 3.5 % was obtained when MLST-I was compared with MLST-II. The internal level of concatenated sequence divergence within MLST-I and MLST-II was 1.4 and 2.7 %, respectively; by comparison with the level of concatenated sequence divergence in established Bcc species, these data demonstrate that the MLST-I and MLST-II lineages represent two distinct species within the Bcc. The latter conclusion was supported by comparison of the whole-genome average nucleotide identity (ANI) level of MLST-I and MLST-II strains with strains of established Bcc species and by a whole-genome-based phylogenetic analysis. We formally propose to classify taxon K bacteria from the MLST-I and MLST-II lineages as Burkholderia contaminans sp. nov. (with strain J2956^T =LMG 23361^T =CCUG 55526^T as the type strain) and Burkholderia lata sp. nov. (with strain 383^T =ATCC 17760^T =LMG 22485^T =CCUG 55525^T as the type strain), respectively. The MLST approach was confirmed as a valuable instrument in polyphasic taxonomic studies; more importantly, the cumulative data for about 1000 Bcc isolates analysed demonstrate that the 3 % concatenated sequence divergence level correlates with the 70 % DNA–DNA hybridization or 95 % whole-genome ANI threshold levels for species delineation.