Several phenomena in astrophysics generate light curves with time delays. Among these are reverberation mapping, and lensed quasars. In some of these systems, the measurement of the time-delay is complicated by the fact that the delayed components are unresolved and that the light curves are generated from a red-noise process. We derive the likelihood function of the observations given a model of either a combination of time-delayed light curves or a single light curve (i.e., the null hypothesis). This likelihood function is different from the auto-correlation function investigated by previous studies. We demonstrate that given a single-band light curve that is a linear combination of two (or more) time-shifted copies of an original light curve, generated from a red-noise probability distribution, it is possible to test if the total-flux light curve is a composition of time-delayed copies or, alternatively, is consistent with being a single copy of the original light curve. We also demonstrate that, in some realistic cases, it is possible to measure the time delays and flux ratios between these unresolved components even when the flux ratio is about 1/10. In the era of synoptic sky surveys, this method is useful for identifying lensed quasars and simultaneously measuring their time delays, and also for estimating the reverberation time scales of active galactic nuclei. In a companion paper, we build on these results to derive another method that uses the center-of-light astrometric position (e.g., of a lensed quasar) along with the combined flux. The combined flux and astrometry allow us to identify lensed quasars and supernovae and measure their time delays, with potentially higher fidelity compared to the flux-only method described in the current work. The astrometry + flux method, however, is not suitable for quasar reverberation mapping. We also comment on the commonly used method of fitting a power-law model to a power spectrum, and present the proper likelihood function for such a fit. We test the new method on simulations and provide Python and MATLAB implementations.