The prognosis of laryngeal squamous cell carcinoma (LSCC) patients remains poor, and early diagnosis can distinctly improve the long-term survival of LSCC patients. MicroRNAs (miRs) are a group of endogenous, noncoding, 18-24 nucleotide length single-strand RNAs and have been demonstrated to regulate the expression of many genes, thus modulating various cellular biological processes. In this study, we aimed to identify critical diagnostic miRNAs based on two machine learning algorithms. The GSE133632 dataset was acquired from the Gene Expression Omnibus (GEO) datasets, comprising LSCC tissular samples (57 specimens) and matched neighboring healthy mucosa tissular samples (57 specimens). Differentially expressed miRNAs (DEMs) were screened between 57 LSCC specimens and 57 normal specimens. The LASSO regression model and SVM-RFE analysis were carried out for the identification of critical miRNAs. ROC assays were applied to evaluate discriminatory ability. We identified 32 DEMs between LSCC specimens and normal specimens. Two machine learning algorithms confirmed that hsa-miR-615-3p, hsa-miR-4652-5p, hsa-miR-450a-5p, hsa-miR-196a-5p, hsa-miR-21-3p, hsa-miR-139-5p, and hsa-miR-424-5p were critical diagnostic factors. According to the ROC assays, seven miRNAs had an AUC value of >0.85 for LSCC. Taken together, our findings identified seven critical miRNAs in LSCC patients which can be used to diagnose LSCC patients with high sensitivity and specificity. These results must be verified by large-scale prospective studies.