Speech production involves a series of processes that operate covertly and cannot be directly accessed. Only their output, speech itself, can be analysed. The collection and investigation of instances of disfluency is an important area of psycholinguistic research since such disharmonic phenomena may yield information concerning the operations that take place in the background: getting to know the causes and courses of speakers committing errors brings us closer to understanding the characteristics of normal processes. In the present study, the authors analyse 1760 segmental-level errors (379 perseverations, 594 anticipations, 401 metatheses, and 386 simple slips of the tongue) and try to find out which level of speech planning is responsible for faulty implementation is each case. What causes errors that can be explained in several different ways? Two assumptions are known. According to one of these, errors can be traced back to planning operations at the phonological level, that is, they precede phonetic planning that is then organised in terms of the faulty phonological pattern. In the other view, impeccable phonological operations are followed by defective phonetic planning that is responsible for the actual errors; in this case, then, we have to do with a simple articulatory implementation problem. The results of the present study confirm the view that segmental-level errors are committed in the phonetic, rather than phonological, planning process. This is connected to the fact that articulatory implementation is the most vulnerable phase of speech production. This may be due to the fact that the least amount of attention is devoted to articulation, this being the most automatic and most well-rehearsed part of the whole process. Further investigations are required to determine the kind of interaction that these operations maintain with the other levels of speech production planning while the individual disfluency phenomena come into being.