Although outbreaks of acute respiratory infection (ARI) at shelters are hypothesized to be associated with shelter crowding, no studies have examined this relationship. We conducted a retrospective study by reviewing medical records of evacuees presenting to one of the 37 clinics at the shelters in Ishinomaki city, Japan, during the 3-week period after the Great Eastern Japan Earthquake and tsunami in 2011. On the basis of a locally weighted scatter-plot smoothing technique, we categorized 37 shelters into crowded (mean space <5·5 m2/per person) and non-crowded (⩾5·5 m2) shelters. Outcomes of interest were the cumulative and daily incidence rate of ARI/10 000 evacuees at each shelter. We found that the crowded shelters had a higher median cumulative incidence rate of ARI [5·4/10 000 person-days, interquartile range (IQR) 0-24·6, P = 0·04] compared to the non-crowded shelters (3·5/10 000 person-days, IQR 0-8·7) using Mann-Whitney U test. Similarly, the crowded shelters had an increased daily incidence rate of ARI of 19·1/10 000 person-days (95% confidence interval 5·9-32·4, P < 0·01) compared to the non-crowded shelters using quasi-least squares method. In sum, shelter crowding was associated with an increased incidence rate of ARI after the natural disaster.