Using a prospective cohort design, we analyzed trends in desire for children and predictors of pregnancy among a cohort of 733 HIV-infected women in rural Uganda who initiated ART between May 2003 and May 2004 and were followed up in their homes until June 2006. Women answered in-depth social and behavioral questionnaires administered every quarter in year 1 after initiating ART, and every 6 to 12 months thereafter. Use of family planning methods was assessed at 18 and 24 months after starting ART. We tested for non-constant pregnancy incidence by using a shape parameter test from the Weibull distribution. We modeled repeated measurements of all variables related to the women's desire for children over time using a generalized estimating equation (GEE) extension to the logistic regression model. Risk factors for pregnancy were examined using Cox proportional hazards model. 711 women eligible for the study were followed-up for a median time of 2.4 years after starting ART. During this time, less than 7% of women reported wanting more children at any time point yet 120 (16.9%) women experienced 140 pregnancies and pregnancy incidence increased from 3.46 per 100 women-years (WY) in the first quarter to 9.5 per 100 WY at 24 months (p<0.0001). This was paralleled by an increase in the proportion of women reporting sexual activity in the past 3 months, from 24.4% at baseline to 32.5% over 24 months of follow-up (p = 0.001). Only 14% of women used permanent or semi-permanent family planning methods by their second year on ART. In the multivariate model, younger age (HR = 2.71 per 10-year decrease, 95% CI: 2.95–3.78), having a BMI>18.5 (HR = 1.09, CI: 1.01–1.18) and not having used condoms consistently in the last 3 months (HR = 1.79, CI: 1.02–3.13) were independently associated with pregnancy.