After a major radioactive incident, accurate dose reconstruction is important for evaluating health risks and appropriate radiation protection policies. After the 2011 Japan Fukushima nuclear incident, we assessed the level of agreement between the modeled and directly measured dose and estimated the uncertainties. The study population comprised 520 school children from Minamisoma city, located 20 km north of the nuclear plant. The annual dose 18–30 months after the incident was assessed using two approaches: estimation using the model proposed by the Japanese government and direct measurement by radiation dosemeters. The ratio of the average of modeled and measured doses was 3.0 (standard deviation (SD): 2.0). The reduction coefficient, an index for radiation attenuation properties, was 0.3 (SD: 0.1) on average, whereas the value used in the government model was 0.6. After adjusting for covariates, the coefficient had a significant negative correlation with the air dose rate in the dwelling location (p < 0.001), indicating that stronger building shielding effects are valuable in areas with higher air contamination levels. The present study demonstrated that some overestimation may have been related to uncertainties in radiation reduction effects, and that the air contamination level might provide a more important indicator of these effects.