In an attempt to explain across-subject variability in fricative spectra, ensemble-averaged spectra were compared to time-averaged spectra for a corpus that consisted of nonsense syllables of the form /pViFV2/ repeated 10-14 times on one breath. Vi and V2 were chosen from /a,i,u/; F was one of /f,v,0,£, s,z,/,2/. Two subjects were recorded saying both this corpus and another corpus consisting of the same fricatives sustained for 3 seconds. A wide variety of articulatory data was also available for these subjects. An ensemble-averaged spectrum could be computed for the beginning (or middle or end) of all fricative tokens. Results show a pattern of spectral change through the fricative that is consistent with aerodynamic and articulatory measures. The non-stationarity thus revealed does not in itself explain the variability across subjects, however. Rather, the ensemble averaging allows a precision in timing of the analysis window which, when coupled with the articulatory data, shows more clearly the effect of vowel context on fricatives, and delineates the differences between /f/ and /$/.