An error analysis was performed on the three ticket vending machines i
nstalled at London underground and overground train stations. A brief
analytic inspection, resulting in a set of predicted errors, was follo
wed by lengthy empirical observations of successes, failures and error
s occurring during machine use. There were two observational phases, 5
years apart. Comparisons were made between the patterns of error-maki
ng on the three machines, using error categories derived from the init
ial analysis. It was found that these comparisons were sufficient to a
ccount for most of the between-machine and between-phase differences,
although some unattributed errors remained. It was also found that muc
h of the observed pattern of error-making had been predicted by the in
itial inspection, and it is suggested that, for relatively simple inte
rfaces such as these, the method (Dialogue Error Analysis) is sufficie
nt to identify and prioritize most problems that will occur in use. At
tempt was also made to relate the observed error categories to influen
tial taxonomies of human error.