For many chaotic systems, accurate calculation of the correlation dimension
from measured data is difficult because of very slow convergence as the sc
ale size is reduced. This problem is often caused by the highly nonuniform
measure on the attractor, This paper proposes a method for collecting data
at large scales and extrapolating to the limit of zero scale. The result is
a vastly reduced required number of data points for a given accuracy in th
e measured dimension. The method is illustrated in detail for one-dimension
al maps and then applied to more complicated maps and flows. Values are giv
en for the correlation dimension of many standard chaotic systems.