The risk/benefit ratio of pharmacotherapy for behavioral symptoms of dementia is questionable: second-generation antipsychotics are poorly tolerated, and the efficacy of alternative treatments, for example, selective serotonin-reuptake inhibitors (SSRIs), is uncertain. Biomarkers of treatment response may improve this risk/benefit ratio. The length polymorphism of the serotonin transporter promoter gene (5-HTTLPR/SLC6A4) may moderate tolerability of SSRIs and expression of behavioral symptoms in dementia. We assessed the effect of 5-HTTLPR on tolerability and efficacy of citalopram and risperidone in a 12-week randomized controlled trial, which included nondepressed patients with dementia hospitalized for behavioral or psychotic symptoms. Genotypes including the A/G polymorphism of the L allele (rs25531) were determined in 92 of 103 participants. We used pattern-mixture models to account for dropout. Low-expression alleles (S and Lg) predicted greater early and overall side effects of citalopram and early treatment discontinuation. These results remained unchanged after excluding African-American participants and in covariate analyses. Unexpectedly, low-expression alleles seemed to predict greater early side effects of risperidone (but not early discontinuation) and poorer early response of psychosis symptoms to risperidone. In conclusion, 5-HTTLPR may be a useful biomarker of SSRI intolerance in dementia. Our findings of intolerance of a second-generation antipsychotics and persistence of psychosis in patients with low-expression alleles needs to be replicated.