In two experiments, the effects of learning on both the accuracy and b
ias of duration judgments were examined. In Experiment 1, subjects lea
rned one of two tasks (i.e., using a computer software package, buildi
ng a model car), containing a varying number of action steps, over a o
ne-, three-, or five-trial period. Retrospective judgments of a task's
total duration revealed that accuracy was high at intermediate stages
of learning but was low at early stages due to an overestimation bias
and low at later stages due to an underestimation bias. The number of
action steps within a task influenced behavior only at early learning
stages where more action steps led to significantly longer duration e
stimates. Experiment 2 acted as a converging operation in which novice
and experienced pianists were asked to estimate, in advance, how long
they thought it would take them to play melodies that varied in their
degree of familiarity (i.e. recently learned, well learned, extremely
well learned). When these estimates were compared with the melodies'
actual playing times, results revealed a similar pattern of accuracy a
nd bias as found in Experiment 1. These findings are discussed in term
s of a ''structural remembering model'' that emphasizes the role of ev
ent predictability in time estimation behavior.