A typical population projection involves performing the arithmetic calculations to project a population from time T to time S (S>T), given a set of assumptions about such population characteristics as fertility, mortality, and migration. These projections are often used as forecasts, yet they are valid forecasts only if the initial set of assumptions hold throughout the projection period. In this paper, we propose a Bayesian population projection technique that incorporates a demographer's uncertainty about the characteristics of a population in the form of elicited prior distributions. This method is particularly useful in populations where census data is sparse or unreliable. We illustrate the methods, from the elicitation of the prior uncertainty to the interpretation of the resulting population distributions, on the Kurdish population of Iraq during the time period 1977-1990.