Abstract.We have observed S4 0954+658 in the B, V, R and I bands for one night in March and two nights in April, 2001, and in the B and I bands for four nights in May, 2002. The observations resulted in almost evenly sampled light curves, 3−8 h long, with an average sampling interval of ∼5−15 min. Because of the dense sampling and the availability of light curves in more than one optical band we are able to study the intra-night flux and spectral variability of the source in detail. Significant observations were observed in all but one case. On average, the flux variability amplitude, on time scales of min/h, increases from ∼2−3% in the I, to ∼3.5−5.5% in the B band light curves. We do not detect any flares within the individual light curves. However, there is a possibility that the April 2001 and late May 2002 observations sample two flares which lasted longer than ∼1−3 days. The evidence is only suggestive though, due to the limited length of the present light curves with regard to the duration of the assumed flares. No spectral variations are detected during the April 2001 observations. The source flux rises and decays with the same rate, in all bands. This variability behaviour is typical of S4 0954+658, and is attributed to geometrical effects. However, significant spectral variations are observed in May 2002. We find that the spectrum hardens/softens as the flux increases/decreases, respectively. Furthermore, the "hardening" rate of the energy spectrum is faster than the rate at which the spectrum becomes "redder" as the flux decays. We also find evidence (although of low statistical significance) that the I band variations are delayed with respect to the B band variations. If the May 2002 observations sample a flaring event, these results suggest that the variations are caused by energetic processes which are associated with the particle cooling and the source light travel time scales.