Metagenomic next-generation sequencing (mNGS) was used to analyze the etiological distribution of refractory pneumonia in children. We compared its efficacy in pathogen diagnosis against traditional methods to provide a basis for clinical adjustment and treatment. A total of 60 children with refractory pneumonia treated at the Department of Respiratory Medicine, Children's Hospital Affiliated with the Capital Institute of Paediatrics, from September 2019 to December 2021 were enrolled in this study. Clinical data (including sex, age, laboratory tests, complications, and discharge diagnosis) and lower respiratory tract specimens were collected, including bronchoalveolar lavage fluid (BALF), deep sputum, pleural effusion, lung abscess puncture fluid, traditional respiratory pathogens (culture, acid-fast staining, polymerase chain reaction, serological testing, etc.), and mNGS detection methods were used to determine the distribution of pathogens in children with refractory pneumonia and to compare the positive rate and diagnostic efficiency of mNGS and traditional pathogen detection for different types of pathogens. Among the 60 children with refractory pneumonia, 43 specimens were positive by mNGS, and 67 strains of pathogens were detected, including 20.90% (14 strains) of which were Mycoplasma pneumoniae, 11.94% (8 strains) were Streptococcus pneumoniae, 7.46% (5 strains) were cytomegalovirus, and 5.97% (4 strains) were Candida albicans. Thirty-nine strains of Mycoplasma pneumoniae (41.03%, 16 strains), Streptococcus pneumoniae (10.26%, 4 strains), Candida albicans (7.69%, 3 strains), and Aspergillus (5.13%, 2 strains) were detected using traditional methods. The positive rate of mNGS detection was 90.48%, and the positive rate of the traditional method was 61.90% (p = 0.050), especially for G+ bacteria. The positive rate of mNGS was greater than that of traditional methods (p < 0.05), but they had no significant difference in detecting G- bacteria, viruses, fungi, or Mycoplasma/Chlamydia. Among the 60 patients, 21 had mixed infections, 25 had single infections, and the other 14 had unknown pathogens. Mycoplasma pneumoniae was most common in both mixed infections and single infections. The sensitivity, specificity, positive predictive value, and negative predictive value of mNGS were 95.45, 37.50, 80.77, and 75.00%, respectively. The sensitivity, specificity, positive predictive value, and negative predictive value of the traditional methods were 72.72, 62.50, 84.21, and 45.45%, respectively. The clinical compliance of mNGS was 80.00%, and that of the traditional method was 70.00%. The sensitivity and negative predictive value of mNGS were high, and the difference in the sensitivity for detecting G+ bacteria was statistically significant (p < 0.05). However, the differences in G- bacteria, fungi, and Mycoplasma/Chlamydia were not statistically significant (p > 0.05). Due to the small sample size, statistical analysis could not be conducted on viral infections. mNGS has higher overall efficacy than traditional methods for the etiological diagnosis of refractory pneumonia in children. The application of mNGS can significantly improve the detection rate of pathogens in children with refractory pneumonia. The sensitivity and negative predictive value of mNGS for detecting G+ bacteria are greater than those of other methods, and it can exclude the original suspected pathogenic bacteria. Unnecessary antibiotic use was reduced, but there was no statistically significant difference in G- bacteria, fungi, or Mycoplasma/Chlamydia.