Measuring time delays between the multiple images of gravitationally lensed quasars is now recognized as a competitive way to constrain the cosmological parameters, and it is complementary with other cosmological probes. This requires long and well sampled optical light curves of numerous lensed quasars, such as those obtained by the COSMOGRAIL collaboration. High-quality data from our monitoring campaign call for novel numerical techniques to robustly measure the delays, as well as the associated random and systematic uncertainties, even in the presence of microlensing variations. We propose three different point estimators to measure time delays, which are explicitly designed to handle light curves with extrinsic variability. These methods share a common formalism, which enables them to process data from n-image lenses. Since the estimators rely on significantly contrasting ideas, we expect them to be sensitive to different bias sources. For each method and data set, we empirically estimate both the precision and accuracy ( bias) of the time delay measurement using simulated light curves with known time delays that closely mimic the observations. Finally, we test the self-consistency of our approach, and we demonstrate that our bias estimation is serviceable. These new methods, including the empirical uncertainty estimator, will represent the standard benchmark for analyzing the COSMOGRAIL light curves.