Confabulation is a mysterious adjunct of amnesia. It remains unexplained why some patients invent untrue stories in response to questions (provoked confabulations) or even spontaneously with no apparent motivation (spontaneous confabulations). Hypothesized mechanisms range from a desire to fill gaps in memory to a loss of the temporal context in memory. We examined the mechanisms of confabulations in 16 amnesic patients. Patients were classified as spontaneous confabulators if they ever acted according to their confabulations. Provoked confabulations were measured as the number of intrusions in a verbal learning test. We found a double dissociation between the two types of confabulations, indicating that they represent different disorders rather than different degrees of the same disorder. Confabulating patients did not show an increased tendency to fill gaps in memory as measured by the number of fake questions concerning non existent items that they answered. Neither type of confabulation correlated with a failure to store new information as gauged with recognition tasks; pure information storage was even found to be normal in some patients. However, we found a positive correlation between several measures of verbal learning and verbal fluency with provoked, but not spontaneous, confabulations. In contrast, spontaneous, but not provoked, confabulations were associated with an inability to recognize the temporal order of stored information as measured by the comparison of two runs of a continuous recognition task. We suggest that provoked confabulations depend on an amnesic subject's search in his deficient memory and are the trade-off for increased item recollection. Spontaneous confabulations appear to be based on a failure to recognize the temporal order of stored information, resulting in erroneous recollection of elements of memory that do not belong together.