An extensive dataset on rice phenology in China, including 202 series broadly covering the past three decades (1980s–2000s), was compiled. From these data, we estimated the responses of growth duration length to temperature using a regression model based on the data with and without detrending. Regression coefficients derived from the detrended data reflect only the temperature effect, whereas those derived from data without detrending represent a combined effect of temperature and confounding cultivar shifts. Results indicate that the regression coefficients calculated from the data with and without detrending show an average shortening of the growth duration of 4.1–4.4 days for each additional increase in temperature over the full growth cycle. Using the detrended data, 95.0% of the data series exhibited a negative correlation between the growth duration length and temperature; this correlation was significant in 61.9% of all of the data series. We then compared the difference between the two regression coefficients calculated from data with and without detrending and found a significantly greater temperature sensitivity using the data without detrending (−2.9 days °C−1) than that derived from the detrended data (−2.0 days °C−1) in the period of emergence to heading for the late rice, producing a negative difference in temperature sensitivity (−0.9 days °C−1). This implies that short-duration cultivars were planted with increase in temperature and exacerbated the undesired phenological change. In contrast, positive differences were detected for the single (0.6 days °C−1) and early rice (0.5 days °C−1) over the full growth cycle, which might indicate that long-duration cultivars were favoured with climate warming, but these differences were insignificant. In summary, our results suggest that a major, temperature induced change in the rice growth duration is underway in China and that using a short-duration cultivar has been accelerating the process for late rice.