The ages of cladogenetic events in Coleoptera are frequently estimated with mitochondrial protein-coding genes (MPCGs) and the "standard" mitochondrial nucleotide substitution rate for arthropods. This rate has been used for different mitochondrial gene combinations and time scales despite it was estimated on short mitochondrial sequences from few comparisons of close related species. These shortcomings may cause greater impact at deep phylogenetic levels as errors in rates and ages increase with branch lengths. We use the full set of MPCGs of 15 species of beetles (two of them newly sequenced here) to estimate the nucleotide evolutionary rates in a reconstructed phylogeny among suborders, paying special attention to the effect of data partitioning and model choices on these estimations. The optimal strategy for nucleotide data, as measured with Bayes factors, was partitioning by codon position. This retrieved Adephaga as a sister group to Myxophaga with strong support (expected-likelihood weights test 0.94-1) and both sisters to Polyphaga, in contradiction with the most currently accepted views. The hypothesis of Archostemata being sister to the remaining Coleoptera, which is in agreement with morphology, was increasingly supported when third codon sites were recoded or completely removed, sequences were analyzed as AA, and heterogeneous models were implemented but the support levels remained low. Nucleotide substitution rates were strongly affected by the choice of data partitioning (codon position versus individual genes), with up to sixfold levels of variation, whereas differences in the molecular clock algorithm produced changes of only about 20%. The global mitochondrial protein coding rate using codon partitioning and an estimated age of 250 million years (MY) for the origin of the Coleoptera was 1.34% per branch per MY, which closely matches the 'standard' clock of 1.15% per MY. The estimation of the rates on alternative topologies gave similar results. Using local molecular clocks, the evolutionary rate in the Polyphaga and Archostemata was estimated to be nearly twice as fast as in the Adephaga and Myxophaga (1.03% versus 0.53% per MY). Rates across individual genes varied from 0.55% to 8.61% per MY. Our results suggest that cox1 might not be an optimal gene for implementing molecular clocks in deep phylogenies for beetles because it shows relatively slow rates at first and second codon positions but very fast rates at third ones. In contrast, nad5, nad4 and nad2 perform better, as they exhibit more homogeneous rates among codon positions. Copyright 2010 Elsevier Inc. All rights reserved.