The mussel Mytilus californianus is a competitive dominant on wave—swept rocky intertidal shores. Mussel beds may exist as extensive monocultures; more often they are an everchanging mosaic of many species which inhabit wave—generated patches or gaps. This paper describes observations and experiments designed to measure the critical parameters of a model of patch birth and death, and to use the model to predict the spatial structure of mussel beds. Most measurements were made at Tatoosh Island, Washington, USA, from 1970—1979. Patch size ranged at birth from a single mussel to 38 m2; the distribution of patch sizes approximates the lognormal. Birth rates varied seasonally and regionally. At Tatoosh the rate of patch formation varied during six winters from 0.4—5.4% of the mussels removed per month. The disturbance regime during the summer and at two mainland sites was 5—10 times less. Annual disturbance patterns tended to be synchronous within 11 sites on one face of Tatoosh over a 10—yr interval, and over larger distances (16 km) along the coastline. The pattern was asynchronous, however, among four Tatoosh localities. Patch birth rate, and mean and maximum size at birth can be used as adequate indices of disturbance. Patch disappearance (death) occurs by three mechanisms. Very small patches disappear almost immediately due to a leaning response of the border mussels (0.2 cm/d). Intermediate—sized patches (<3.0 m2) are eventually obliterated by lateral movement of the peripheral mussels: estimates based on 94 experimental patches yield a mean shrinking rate of 0.05 cm/d from each of two principal dimensions. Depth of the adjacent mussel bed accounts for much of the local variation in closing rate. In very large patches, mussels must recruit as larvae from the plankton. Recovery begins at an average patch age of 26 mo; rate of space occupation, primarily due to individual growth, is 2.0—2.5%/mo. Winter birth rates suggest a mean turnover time (rotation period) for mussel beds varying from 8.1—34.7 yr, depending on the location. The minimal value is in close agreement with both observed and calculated minimal recovery times. Projections of total patch area, based on the model, are accurate to within 5% of the observed. Using a method for determining the age of patches, based on a growth curve of the barnacle Balanus cariosus, the model permits predictions of the age—size structure of the patch population. The model predicts with excellent resolution the distribution of patch area in relation to time since last disturbance. The most detailed models which include size structure within age categories are inconclusive due to small sample size. Predictions are food for large patches, the major determinants of environmental patterns, but cannot deal adequately with smaller patches because of stochastic effects. Colonization data are given in relation to patch age, size and intertidal position. We suggest that the reproductive season of certain long—lived, patch—dependent species is moulded by the disturbance regime. The necessary and vital connection between disturbance which generates spatial pattern and species richness in communities open to invasion is discussed.