We studied the time-space variations of mortality for parkinsonism in the US during the periods 1962-1985 and 1971-1978 from statistics on primary cause-of-death and multiple causes-of-death, respectively. Linear regression analysis and a test for significance of spatial clustering were used. For parkinsonism as a primary cause-of-death, up to the late 1970's there was a decrease in mortality for the age groups below 75 and stable rates for the age group 75 years and over. A moderate decrease among those below 75 years, and a sharp increase in mortality among those over 74, were observed for the period since the late 1970s. The variation of the age-specific mortality during the period 1962-1985 ranged from 100% in the age group 35-44 years to +98% in those aged over 84 years. Within the same age group, the variations across time of mortality due to, and related to parkinsonism, for the period 1971-1978, were similar. In the spatial analysis, an association between mortality related to parkinsonism and geographical latitude was found. We conclude that: 1) parkinsonism at death is widely distributed; 2) the changes across time can not be explained by a reduction in general mortality; 3) improvements in diagnostic ascertainment and reporting among the elderly may have particularly occurred during the last decade; and 4) the disorder has been progressively confined to the elderly by diminishing in the younger birth cohorts.