Invasive bacterial and candidal infections are known to involve the retina, but the natural history of the retinal lesions and the utility of ophthalmologic consultation in the critical care setting as a diagnostic tool are not well understood. We 1) performed weekly funduscopic examinations on 77 medical and surgical patients in intensive care units (ICUs), 2) analyzed results of serial ocular examinations in 180 non-neutropenic patients with candidemia, and 3) reviewed the English literature on the association of retinal lesions with disseminated bacterial or candidal infection (DBCI). We found that 15 (19%) of the ICU patients had retinal lesions consistent with DBCI. Of these 15, 1 had clearly sepsis-related retinal lesions, while 13 (87%) had 1 or more systemic disease that could have explained their retinal findings (6 diabetic retinopathy; 2 human immunodeficiency virus infection/acquired immunodeficiency syndrome (HIV/AIDS) retinopathy; 2 hypertensive retinopathy; 1 hemolytic uremic syndrome, and 1 leukemia). Multivariate analysis revealed that systemic disease (odds ratio 8.37, 95% confidence intervals: 3.24-21.56) independently correlated with the presence of retinal lesions while DBCI, trauma, hyperalimentation, and transfusion of blood products were not independently predictive in any analysis. Twenty of the 180 (15%) candidemic patients had retinal lesions. Two (1%) had classic 3-dimensional white lesions with vitreal extension, and 5 (2.7%) had chorioretinal lesions without vitreal haziness. Notably, 10% of patients had superficial retinal hemorrhages and/or cotton wool spots that could have been due to either candidemia or a systemic disease (diabetes, hypertension, renal failure, closed head trauma). Concurrent bacteremia occurred in 3 of the 27 patients with eye lesions. Retinal lesions resolved in a mean of 33 days. None of the patients had symptoms at the time of the retinal finding. We found 3 studies that prospectively assessed retinal lesions in bacteremic patients. The frequency of retinal lesions in these series varied from 12% to 26%, with the most common lesions being cotton wool spots followed by superficial retinal hemorrhages. White-centered hemorrhages were seen in about 15% +/- 2 of bacteremic patients. Five studies prospectively evaluated candidemic patients for Candida endophthalmitis. These studies observed rates from 0% to 78% for lesions consistent with candidal endophthalmitis. Most studies performed recently found that nonspecific lesions such as cotton wool spots or superficial retinal hemorrhages occurred with a frequency of 11% to 20%. The availability of less toxic antifungal agents, more frequent use of empirical therapy, and the trend to early treatment may be altering the frequency of this complication. Observation of a classic 3-dimensional retina-based vitreal inflammatory process is virtually diagnostic of endogenous endophthalmitis due to Candida spp., but such lesions are relatively uncommon. Conversely, nonspecific lesions that could be due to bacterial or candidal endophthalmitis (cotton wool spots, retinal hemorrhages, and Roth spots) are seen frequently. These lesions are most often due to an underlying systemic disease rather than an infection. Serial examinations provide the best evidence that a given lesion is due to an intercurrent infection. The current low rate of vitreal extension of retinal process appears to be due to the high rate of empirical or therapeutic use of antifungal agents in high-risk patient groups. Ophthalmoscopy should be performed in patients with known candidemia. However, ophthalmoscopic examination seems to have little value in assisting with the discovery of occult disseminated candidiasis or bacterial infection.