The question that is addressed in this contribution is: to what extent does the language production mechanism in children differ from that in competent adult speakers? It is assumed that utterance planning in adults is a hierarchical process, in which relatively autonomous processing components successively build conceptual, syntactic and phonological representations. Each component of the production system, however, makes use of the same basic processing algorithm, the so-called frame-and-filler mechanism. The developmental question can thus be rephrased as follows: (a) Do the representations constructed at the various levels change as a function of age? (b) Do the basic planning algorithms change? Three longitudinal studies of spontaneous speech production in 1.5 to 4-year-old children are reviewed. The first study presented a child in which the transition from telegraphic speech to a morpho-syntactically more mature type of language use led to a temporary increase in speech disfluency (interruptions, repetitions, self-corrections etc.). It is argued that the disfluency peak reflects a re-arrangement of syntactic planning, which is necessitated by the development of the child's morpho-syntactic knowledge. The second study compared slips of the tongue in adults and children, particularly word and sound substitutions. Basically, the patterns found were similar for adults and children. Not only in the adults, but also in the children the linguistic elements that interact in speech errors are inherently or contextually similar. This suggests that the frame-and-filler mechanism is developmentally invariant. The third study investigated the optional realization of metrically weak syllables and morphemes in children's utterances. The results indicate that this phenomenon can be related to a constraint on phonological planning frames, viz. that only trochaeic feet are allowed. The conclusion that can be drawn from this review is that in the development of language production, the representations (planning frames) change, but the planning algorithms remain essentially the same.