CD33 is variably expressed on leukemia blasts in almost all patients with acute myeloid leukemia (AML) and possibly leukemia stem cells in some. Efforts to target CD33 therapeutically have focused on gemtuzumab ozogamicin (GO; Mylotarg), an antibody-drug conjugate delivering a DNA-damaging calicheamicin derivative. GO is most effective in acute promyelocytic leukemia but induces remissions in other AML types and received accelerated approval in the US in 2000. However, because a large follow-up study showed no survival improvement and increased early deaths the drug manufacturer voluntarily withdrew the US New Drug Application in 2010. More recently, a meta-analysis of data from several trials reported better survival in adults with favorable- and intermediate-risk cytogenetics but not adverse-risk AML randomized to receive GO along with intensive induction chemotherapy. As a result, GO is being re-evaluated by regulatory agencies. Responses to GO are diverse and predictive biological response markers are needed. Besides cytogenetic risk, ATP-binding cassette transporter activity and possibly CD33 display on AML blasts may predict response, but established clinical assays and prospective validation are lacking. Single-nucleotide polymorphisms in CD33 may also be predictive, most notably rs12459419 where the minor T-allele leads to decreased display of full-length CD33 and preferential translation of a splice variant not recognized by GO. Data from retrospective analyses suggest only patients with the rs12459419 CC genotype may benefit from GO therapy but confirmation is needed. Most important may be markers for AML cell sensitivity to calicheamicin, which varies over 100 000-fold, but useful assays are unavailable. Novel CD33-targeted drugs may overcome some of GO's limitations but it is currently unknown whether such drugs will be more effective in patients benefitting from GO and/or improve outcomes in patients not benefitting from GO, and what the supportive care requirements will be to enable their safe use.