Cardiogenic shock (CS) is a life-threatening emergency. New biomarkers are needed in order to detect patients at greater risk of adverse outcome. Our aim was to assess the characteristics of miR-21-5p, miR-122-5p, and miR-320a-3p in CS and evaluate the value of their expression levels in risk prediction. Circulating levels of miR-21-5p, miR-122-5p, and miR-320a-3p were measured from serial plasma samples of 179 patients during the first 5–10 days after detection of CS, derived from the CardShock study. Acute coronary syndrome was the most common cause (80%) of CS. Baseline (0 h) levels of miR-21-5p, miR-122-5p, and miR-320a-3p were all significantly elevated in nonsurvivors compared to survivors (p < 0.05 for all). Above median levels at 0h of each miRNA were each significantly associated with higher lactate and alanine aminotransferase levels and decreased glomerular filtration rates. After adjusting the multivariate regression analysis with established CS risk factors, miR-21-5p and miR-320a-3p levels above median at 0 h were independently associated with 90-day all-cause mortality (adjusted hazard ratio 1.8 (95% confidence interval 1.1–3.0), p = 0.018; adjusted hazard ratio 1.9 (95% confidence interval 1.2–3.2), p = 0.009, respectively). In conclusion, circulating plasma levels of miR-21-5p, miR-122-5p, and miR-320a-3p at baseline were all elevated in nonsurvivors of CS and associated with markers of hypoperfusion. Above median levels of miR-21-5p and miR-320a-3p at baseline appear to independently predict 90-day all-cause mortality. This indicates the potential of miRNAs as biomarkers for risk assessment in cardiogenic shock.