Prediction of return flow moisture over the Gulf of Mexico has been a continual problem for operational numerical weather prediction. Part of the problem resides in the uncertainty in the initial state (especially over the Gulf), uncertainty in the boundary conditions (especially the sea surface temperature), and uncertainty in the model itself (inability to accurately account for the physical processes at the sea/air interface such as turbulent flux of heat and moisture). These uncertainties naturally contribute to the uncertainty in the forecast. In an effort to understand the problem of predicting moisture in these return flow situations, a Lagrangian mixed-layer model is used. This model has approximately 25 elements in its control vector (initial condition, boundary condition, and physical parameters). Data collected during GUFMEX (1988) is used in conjunction with the model. A control state, i. e., a basic state, is found that is most faithful to the observations. We then explore the stochastic-dynamic prediction of the mixed layer properties (height of layer, temperature of layer, moisture in layer, jumps in temperature and moisture atop the mixed layer) by perturbing the control vector. A Monte Carlo approach is used with 1000 member samples to construct the frequency distribution of the forecast variables at t= 18 h. Analytic solution of the model equations under special conditions allows us to validate the finite-difference solution and derivatives of the solution with respect to the elements of the control vector. The goodness of fit between analytic and finite-difference forms of solution and derivative calculation lend credibility to the results that stem from the more-complicated form of the model. The primary results from the research are: 1) analytic solution indicates that the change in the forecast variables at t=18 h are well-approximated by the Taylor expansion to first derivative only — in short, an adjoint-derived sensitivity based only on the first derivatives would give an accurate result, 2) uncertainty in the forecast of the moisture, mixed-layer height, and temperature of the mixed-layer, are less sensitive to initial condition uncertainty than to uncertainty in the sea surface temperature (boundary condition) and physical parameters (turbulent flux coefficients and entrainment parameter), 3) the model sensitivity based on the so-called “optimal perturbation”, an approach commonly used in adjoint model sensitivity studies, gives an unrealistically large change in the forecast — i. e., the optimal perturbation method gives a change in the output that is in the far wings of the frequency distribution (much greater that the standard deviation), 4) results indicate that improvements in the parameterization of the turbulent fluxes at the sea/air boundary and accuracy of sea surface temperatures should go a long way toward improving the prediction of moisture in return flow events over the Gulf of Mexico.