This work describes a hydrogel fluorescence microsensor for prolonged stable temperature measurements. Temperature measurement using microsensors has the potential to provide information about cells, tissues, and the culture environment, with optical measurement using a fluorescent dye being a promising microsensing approach. However, it is challenging to achieve stable measurements over prolonged periods with conventional measurement methods based on the fluorescence intensity of fluorescent dye because the excited fluorescent dye molecules are bleached by the exposure to light. The decrease in fluorescence intensity induced by photobleaching causes measurement errors. In this work, a photobleaching compensation method based on the diffusion of fluorescent dye inside a hydrogel microsensor is proposed. The factors that influence compensation in the hydrogel microsensor system are the interval time between measurements, material, concentration of photo initiator, and the composition of the fluorescence microsensor. These factors were evaluated by comparing a polystyrene fluorescence microsensor and a hydrogel fluorescence microsensor, both with diameters of 20 µm. The hydrogel fluorescence microsensor made from 9% poly (ethylene glycol) diacrylate (PEGDA) 575 and 2% photo initiator showed excellent fluorescence intensity stability after exposure (standard deviation of difference from initial fluorescence after 100 measurement repetitions: within 1%). The effect of microsensor size on the stability of the fluorescence intensity was also evaluated. The hydrogel fluorescence microsensors, with sizes greater than the measurement area determined by the axial resolution of the confocal microscope, showed a small decrease in fluorescence intensity, within 3%, after 900 measurement repetitions. The temperature of deionized water in a microchamber was measured for 5400 s using both a thermopile and the hydrogel fluorescence microsensor. The results showed that the maximum error and standard deviation of error between these two sensors were 0.5 °C and 0.3 °C, respectively, confirming the effectiveness of the proposed method.