The triggering receptor expressed on myeloid 2 (TREM2) is an immune phagocytic receptor expressed on brain microglia known to trigger phagocytosis and regulate the inflammatory response. Homozygous mutations in TREM2 cause Nasu–Hakola disease, a rare recessive form of dementia. A heterozygous TREM2 variant, p.R47H, was recently shown to increase Alzheimer’'s disease (AD) risk. We hypothesized that if TREM2 is truly an AD risk gene, there would be additional rare variants in TREM2 that substantially affect AD risk. To test this hypothesis, we performed pooled sequencing of TREM2 coding regions in 2082 AD cases and 1648 cognitively normal elderly controls of European American descent. We identified 16 non-synonymous variants, six of which were not identified in previous AD studies. Two variants, p.R47H [P = 9.17 × 10⁻⁴, odds ratio (OR) = 2.63 (1.44–4.81)] and p.R62H [P = 2.36 × 10⁻⁴, OR = 2.36 (1.47–3.80)] were significantly associated with disease risk in single-variant analyses. Gene-based tests demonstrate variants in TREM2 are genome-wide significantly associated with AD [P_SKAT-O = 5.37 × 10⁻⁷; OR = 2.55 (1.80–3.67)]. The association of TREM2 variants with AD is still highly significant after excluding p.R47H [P_SKAT-O = 7.72 × 10⁻⁵; OR = 2.47 (1.62–3.87)], indicating that additional TREM2 variants affect AD risk. Genotyping in available family members of probands suggested that p.R47H (P = 4.65 × 10⁻²) and p.R62H (P = 6.87 × 10⁻³) were more frequently seen in AD cases versus controls within these families. Gel electrophoresis analysis confirms that at least three TREM2 transcripts are expressed in human brains, including one encoding a soluble form of TREM2.