Statistical analysis of strong-motion records of the 1966 Parkfield earthquake and the 1952 Taft earthquake indicates that the usable limit of long period of velocity and displacement, calculated by integration of the accelerogram records, is restricted mainly by human reading and base line correction errors. Base line correction errors arise from uncertainties in the coefficients of the fitted straight base line. Calculation by the jackknife method of the variances of the errors in displacement records from linear base line corrections indicates that these errors are more pronounced in the long-period (above 7 sec) range than errors from digitization. The records analyzed in this paper indicate that long-period limits due to the combined errors vary between 7 and 14 sec. Beyond these limits components of displacement spectra from the present analog accelerograms are not reliable measures of ground motion.