Loudness has been estimated traditionally based on spectral properties of the sound and critical band analysis of the ear: Spectral components add energy within the critical band and add loudness across critical bands. This traditional model is successful in predicting loudness of steady-state sounds but has not been verified with dynamic sounds like speech. This study conducted three experiments using modern loudness balance procedures to study temporal and spectral factors in loudness sensation. Experiment 1 repeated Zwicker's two-tone and three-tone complex experiment in which frequency and phase differences between tones were varied systematically [Zwicker and Fastl, Psychoacoustics (Springer-Verlag, 1990), pp. 189--190]. While the results generally were consistent with the traditional model, a deviation occurred for the frequency difference less than 10 Hz, where loudness was affected by the low-frequency temporal envelope and was greater than that predicted by the energy summation model. To further separate spectral from temporal factors and to rule out the influence of critical bands, experiment 2 used sinusoidally amplitude-modulated noises in normal-hearing listeners and experiment 3 used cochlear implant listeners. Data from experiments 2 and 3 indicated that the temporal envelope in modulated stimuli can increase loudness by 3 dB compared with unmodulated stimuli of equal rms level.