This study addresses the long-standing uncertainty about the internal classification of soldier beetles (Elateroidea: Cantharidae). Four datasets were compiled and analysed: 66 genes for 14 terminals, 15 mtDNA genes for 79 terminals, one mtDNA and two rRNA genes for 217 terminals, and barcodes for 576 terminals. Based on congruent topologies, Chauliognathinae is proposed as a sister to the remaining Cantharidae, followed by the redefined Malthininae (including Tytthonyxini), the paraphyletic "dysmorphocerine" lineages (Dysmorphocerinae sensu stricto and Heteromastiginae subfam. nov.), and Silinae + Cantharinae as a terminal clade. The present phylogeny supersedes earlier morphology and short-fragment molecular hypotheses that have not converged on a consensus. Few morphological characters corroborate the DNA-based relationships (see the adults and larval keys). However, morphology-based hypotheses have relied on a few informative characters, and no evidence strongly rejects the preferred molecular topology. The interpretation of morphological characters and uncertain polarity resulting from the high phenotypic disparity of Elateroidea are discussed in detail. The dated phylogeny hypothesizes the earliest split within the Cantharidae in the Berriasian stage (Early Cretaceous, ~141 Myr) and the diversification of most extant subfamilies and tribes already in the Late Cretaceous. The most diverse subfamily, Cantharinae, represents a delayed radiation that started during the Eocene climatic optimum, 55.5 Myr. The late origin of Cantharinae questions the classification of Cretaceous Cantharidae as members of Cantharinae. Instead, the results suggest their deeper rooting after separating from dysmorphocerine lineages and before the node between Cantharinae and Silinae.