Mutation is a phenomenon inescapable for all life-forms, including bacteria. While bacterial mutation rates are generally low due to the operation of error-avoidance systems, sometimes they are elevated by many orders of magnitude. Such a state, known as a hypermutable state, can result from exposure to stress or to harmful environments. Studies of bacterial mutation frequencies and analysis of the precise types of mutations can provide insights into the mechanisms by which mutations occur and the possible involvement of error-avoidance pathways. Several approaches have been used for this, like reporter assays involving non-essential genes or mutation accumulation over multiple generations. However, these approaches give an indirect estimation, and a more direct approach for determining mutations is desirable. With the recent development of a DNA sequencing technique known as Duplex Sequencing, it is possible to detect rare variants in a population at a frequency of 1 in 107 base pairs or less. Here, we have applied Duplex Sequencing to study spontaneous mutations in E. coli. We also investigated the production of replication errors by using a mismatch-repair defective (mutL) strain as well as oxidative-stress associated mutations using a mutT-defective strain. For DNA from a wild-type strain we obtained mutant frequencies in the range of 10-7 to 10-8 depending on the specific base-pair substitution, but we argue that these mutants merely represent a background of the system, rather than mutations that occurred in vivo. In contrast, bona-fide in vivo mutations were identified for DNA from both the mutL and mutT strains, as indicated by specific increases in base substitutions that are fully consistent with their established in vivo roles. Notably, the data reproduce the specific context effects of in vivo mutations as well as the leading vs. lagging strand bias among DNA replication errors.