The diagnosis of primary central nervous system (CNS) lymphoma, which is predominantly of the diffuse large B-cell lymphoma type (CNS DLBCL), is challenging. MicroRNAs (miRs) are gene expression-regulating non-coding RNAs that are potential biomarkers. We aimed to distinguish miR expression patterns differentiating CNS DLBCL and non-malignant CNS diseases with tumor presentation (n-ML). Next generation sequencing-based miR profiling of cerebrospinal fluids (CSFs) and brain tumors was performed. Sample source-specific (CSF vs. brain tumor) miR patterns were revealed. Even so, a set of 17 miRs differentiating CNS DLBCL from n-ML, no matter if assessed in CSF or in a tumor, was identified. Along with the results of pathway analyses, this suggests their pathogenic role in CNS DLBCL. A combination of just four of those miRs (miR-16-5p, miR-21-5p, miR-92a-3p, and miR-423-5p), assessed in CSFs, discriminated CNS DLBCL from n-ML samples with 100% specificity and 67.0% sensitivity. Analyses of paired CSF-tumor samples from patients with CNS DLBCL showed significantly lower CSF levels of miR-26a, and higher CSF levels of miR-15a-5p, miR-15b-5p, miR-19a-3p, miR-106b-3p, miR-221-3p, and miR-423-5p. Noteworthy, the same miRs belonged to the abovementioned set differentiating CNS DLBCL from non-malignant CNS diseases. Our results not only add to the basic knowledge, but also hold significant translational potential.