The calculation of finite temperature electron spin resonance (ESR) spectra for concrete specified crystal configurations is a very important issue in the study of quantum spin systems. Although direct evaluation of the Kubo formula by means of numerical diagonalization yields exact results, memory and CPU time restrictions limit the applicability of this approach to small system sizes. Methods based on the time evolution of a single pure quantum state can be used to study larger systems. One such method exploits the property that the expectation value of the autocorrelation function obtained for a few samples of so-called thermal typical states yields a good estimate of the thermal equilibrium value. In this paper, we propose a new method based on a Wiener-Khinchin-like theorem for quantum system. By comparison with exact diagonalization results, it is shown that both methods yield correct results. As the Wiener-Khinchin-based method involves sampling over thermal typical states, we study the statistical properties of the sampling distribution. Effects due to finite observation time are investigated and found to differ for the two methods but it is also found that, for both methods, the effects vanish as the system size increases. We present ESR spectra of the one-dimensional XXZ Heisenberg chain of up to 28 spins and discuss the dependence of separation of double peaks on the chain length.