Creatinine level in body fluid has been most used for early diagnosis of kidney function and renal diseases; therefore, it is desirable to develop a rapid and quantitative method for creatinine detection. Herein, surface-enhanced Raman spectroscopy (SERS) based on evaporation-induced optimal hotspots of Au nanocubes has been employed to quantitatively detect creatinine in human serum. Time-dependent SERS measurements during the evaporating process clearly prove that the optimal hotspots are formed on the eve of droplet drying of a Au sol and analyte mixture. The reproducibility tested by independent droplet experiments shows a relative standard deviation of SERS intensity of about 4.16%. Furthermore, the quantitative analysis of creatinine shows a clear linear fitting between the logarithmic SERS intensity and the creatinine concentrations with a correlation coefficient of 0.987. The creatinine level in the serum from volunteers estimated by this method shows an average difference of less than 5% compared to that by the clinical Jaffe reaction method. The evaporation-induced optimal hotspot method with good reproducibility offers a simple and feasible way for practical SERS detection.