This paper investigates whether non-linear gravitational instability c
an account for the clustering of galaxies on large and small scales, a
nd for the evolution of clustering with epoch. The local clustering sp
ectrum is accurately established, and bias is not a great source of un
certainty: the real-space power spectra of optical and IRAS galaxies s
how only a weak scale-dependent relative bias of b similar or equal to
1.15 on large scales, increasing to b similar or equal to 1.5 on the
smallest scales. Comparison with results in redshift space favours a r
elatively small distortion parameter beta(opt) = Omega(0.6)/b(opt) sim
ilar or equal to 0.4. No CDM-like spectrum is consistent with the shap
e of the observed non-linear spectrum. Unbiased low-density models gre
atly overpredict the small-scale correlations; high-density models wou
ld require a bias which does not vary monotonically with scale. The tr
ue linear power spectrum contains a primordial feature at k similar or
equal to 0.1 h Mpc(-1), and must break quite abruptly to an effective
slope of n less than or similar to -2.3 on smaller scales. This empir
ical fluctuation spectrum also fits the CFRS data on the evolution of
clustering, provided that the Universe is open with Omega similar or e
qual to 0.3. Only this case explains naturally how the small-scale spe
ctrum can evolve at the observed rate while retaining the same power-l
aw index. An unbiased open model also matches correctly the large-scal
e COBE data, and offers an attractively simple picture for the phenome
nology of galaxy clustering.