Relaxation times estimation methods play a central role in various problems, such as magnetic resonance (MR) hardware calibration, tissue characterization, or temperature measurement. Previous studies have proposed optimization criteria to estimate the relaxation time T1 faster than with a multipoint method leading to two-point methods. In this paper, the class of optimized two-point methods is extended to gradient-echo (GE) sequence offering new advantages over spin-echo (SE) or inversion recovery (IR) sequences. Two GE acquisitions, with optimal flip angles theta1 and theta2 minimizing both the total scan time and the variance in the computed T1 image were applied to estimate T1, and the results were compared with those of SE sequence with optimized paired repetition times T(R1) and T(R2). First, phantom studies were carried out with five tissue-like samples on a 0.5T scanner. Then in vivo, human brain T1 image were calculated using both optimized GE and SE two-point methods. More precise T1 GE estimates than those for SE were found thanks to high signal-to-noise ratio (SNR) per unit of time, but with a small bias. These results also concern the temperature variation measurement methods, based on T1 estimation. Preliminary experimental data for temperature measurement are given.