Four experiments were conducted to explore the timbres of natural, continua
nt orchestral instruments with emulation based on sampling, frequency modul
ation (FM) synthesis, and a hybrid consisting of sampling and synthesis tec
hniques combined. Identification of instruments using verbal labels was sig
nificantly better for the natural and sampling-based signals than for eithe
r FM synthesis or the hybrid technique, a result also found for aural categ
orization. Perceptual scaling of timbral similarities demonstrated great co
nsistency across a series of independent variables, including musical train
ing, monophonic and stereo presentation, and long versus short signal durat
ions. The first dimension of the classical multidimensional scaling (CMDS)
solutions mapped onto long-time-average spectral centroid. The second dimen
sion mapped onto a measure of spectral variability. Little evidence was fou
nd for the mapping of attack time or signal duration onto either dimension.
A third dimension separated most natural instruments from their emulated c
ounterparts. Experiments using verbal attribute ratings confirmed the corre
lation of spectral centroid, the first dimension of the perceptual space, a
nd ratings of nasality; the second dimension correlated with spectral varia
bility and modestly correlated with ratings of rich, brilliant, and tremulo
us. Mismatches of spectral distribution and variability resulted in poor em
ulations of the natural instruments. Results suggest that further study of
centroid and time-variant psychophysical properties is warranted.