<b><i>Background:</i></b> The clinical challenge in subjective cognitive impairment (SCI) is to identify which individuals will present cognitive decline. We created a statistical model to determine which variables contribute to SCI and mild cognitive impairment (MCI) versus Alzheimer's disease (AD) diagnoses. <b><i>Methods:</i></b> A total of 993 subjects diagnosed at a memory clinic (2007-2009) were included retrospectively: 433 with SCI, 373 with MCI and 187 with AD. Descriptive statistics were provided. A logistic regression model analyzed the likelihood of SCI and MCI patients being diagnosed with AD, using age, gender, Mini-Mental State Examination score, the ratio of β-amyloid 42 divided by total tau, and phosphorylated tau as independent variables. <b><i>Results:</i></b> The SCI subjects were younger (57.8 ± 8 years) than the MCI (64.2 ± 10.6 years) and AD subjects (70.1 ± 9.7 years). They were more educated, had less medial temporal lobe atrophy (MTA) and frequently normal cerebrospinal fluid biomarkers. Apolipoprotein E4/E4 homozygotes and apolipoprotein E3/E4 heterozygotes were significantly less frequent in the SCI group (6 and 36%) than in the AD group (28 and 51%). Within the regression model, cardiovascular risk factors, confluent white matter lesions, MTA and central atrophy increased the AD likelihood for SCI subjects. <b><i>Conclusions:</i></b> SCI patients form a distinct group. In our model, factors suggesting cardiovascular risk, MTA and central atrophy increased the AD likelihood for SCI subjects.