The transmission of a quasar spectrum (flux divided by the continuum) is correlated because of the finite width of absorption lines. We describe a technique for calculating the transmission correlation function produced by randomly distributed lines. We also introduce straightforward procedure for measuring the pixel-pixel transmission correlation function xi(pp) directly from observed quasar spectra. We apply the method to 12 Sargent, Boksenberg, & Steidel QSO spectra and compare these with theoretical transmission correlation functions and with xi(pp) measured from computer-simulated quasar spectra of Lyalpha forest models with Poisson-distributed lines. The simulations are designed to mimic the observed spectrum as closely as possible, with the same wavelength sampling, instrumental resolution, continuum and noise properties. The comparisons with line distributions that are power laws in column density and redshift, and Gaussians in line width b reveal an excess in the observed xi(pp) at DELTAv congruent-to 150 km s-1, if we adopt the Carswell et al. (1991) parameters for the Gaussian (mean bo = 30 km s-1, dispersion sigma(b) = 10 km s-1). One possibility is that the Lyalpha forest lines are actually clustered at velocity separation scales DELTAv congruent-to 150 km s-1. Another possibility we explore here is that the b-distribution has more large b clouds and a larger dispersion. We find the observed xi(pp) is barely consistent with b0 = 40 km s-1 and sigma(b) = 25 km s-1. We show that the measured xi(pp) is relatively insensitive to the noise level and to errors in the continuum determination, unlike the traditional line counting methods, where the outcome is quite vulnerable to both. It also requires no line deblending and thus offers a powerful tool for extracting information from the crowded Lyalpha forest.