The accentual patterns in Japanese can be consistently described in terms of rapid rises and falls in pitch, and their timing can be quantitatively specified by the onset and offset of the underlying accent command applied to a model for F[inf 0] contour generation [H. Fujisaki and K. Hirose, J. Acoust. Soc. Jpn. (E) 5, 233--242 (1984)]. The present paper describes a quantitative study on the timing of accent commands relative to the acoustic-phonetic characteristics of the morae and phones in utterances of the common Japanese. The speech material consists of words of varying length, phonetic constituents, and accent types, embedded in a carrier sentence. The timing of the accent command was estimated from the F[inf 0] contour by analysis-by-synthesis, while the timing of phones and morae was detected by referring to the frequency-time-intensity patterns. It was found that the relative timing of the accent command can be consistently described in reference to the onset of vowel rather than the onset of a mora when the latter starts with a consonant. Systematic variations in the timing due to phonetic constituency and accent type are formulated as rules for speech synthesis.