BACKGROUND AND AIM: Missing data in air pollutant monitoring is a common problem observed in urban air quality networks that compromises the internal validity of epidemiological studies and air quality decision making. AIM: To evaluate imputation methods to replace missing data in long-term exposure. METHOD: Ecological study, with data the Air Monitoring Network of Bogotá D.C., Colombia for nitrogen dioxide during 2018. Several imputation techniques were evaluated: mean, median, random forest, decision forest, classification trees using regressions, predictive mean matching, linear regression and neural network). Measures of central tendency, dispersion and line graphs were used to compare methods RESULTS: Missing data were assumed to follow a completely random pattern (Missing Completely at Random), since the loss was due to a power outage. The percentage of missing data at the Carvajal Sevillana (15%), Centro de Alto Rendimiento (19%), Guaymaral (21%) and Puente Aranda (8%) stations was less than 25%. For the mean and median imputation techniques, similar results were obtained in the summary measures, differing with the predictive values of the linear regression. On the other hand, the techniques of classification trees by regression, random forests and decision forests, underestimate the values of the summary measures with respect to the values without imputation. While imputation with predictive mean adjustment and neural network, similar values were observed with the data observed without imputation CONCLUSIONS: In conclusion, it was observed that the imputation with random forest, predictive mean adjustment and neural network provides similar data to those observed for 24-hour mean nitrogen dioxide concentrations.