Efficient and precise yield prediction is critical to optimize celery(Apium graveolens L.) yield and guide successful agricultural decision making. In this study，a two-year field experiment was conducted to establish an in-season and accurate yield prediction model for celery by using the GreenSeeker hand-hold optical sensor. Two different cultivars Xiqing(XQ)and Yaoqing(YQ)were selected. Normalized difference vegetation index(NDVI)was measured at different stages with the model establishing and validating to test the feasibility and accuracy of GreenSeeker sensor-based yield prediction model. Pearson correlation analysis was performed to identify the relationship between the NDVI measurements and the harvested yield of celery，and the correlation coefficient was observed from 0.583 to 0.805 for XQ and from 0.256 to 0.922 for YQ. Regression analysis showed that the NDVI measurements at 30~70 d after transplanting could predict in-season yield of celery accurately，with R2 values ranged from 0.51 to 0.85，and no significant difference existed between different predict models. Combining the results of different cultivars，the canopy NDVI measurements could predict the in-season yield of XQ and YQ，and the R2 and RMSE values were ranged from 0.66 to 0.84 and from 7.40 to 10.72，respectively. It was verified that the slopes of the validation equations were 0.95 and 1.05 at 60 and 70 d after transplanting，respectively. The data points of predicted yield and measured yield were evenly distributed around the 1∶1 line，indicating that the outer leaf flourish stage was the earliest stage for yield prediction with the canopy sensor. Therefore，establishing a yield prediction model based on NDVI measurements can accurately reflect the in-season yield of celery and provide theoretical basis for the yield prediction and precise nutrient management of celery and other vegetable crops.