The applicability of a four-dimensional variational data assimilation (4DVAR) technique to retrieval of microscale turbulent structures in a convective boundary layer is assessed. Two new features are implemented into the existing 4DVAR model: a height-dependent eddy viscosity and a surface flux model. The identical twin experiments approach is adopted to utilize the model itself to generate 13 instantaneous three-dimensional radial velocity datasets uniformly spanning 5 min. An ideal experiment, using these datasets as the observations, is first tested. After 400 iterations, the resulting correlation coefficients between retrieved and exact data are 0.99 for velocity and 0.97 for temperature fields. To emulate the lidar scanning feature, the 13 three-dimensional datasets are used to construct two volume scan datasets with each horizontal data slice taken from different instantaneous datasets. Using these data as the input, the correlation coefficients for horizontal, spanwise, and vertical velocity fluctuations and temperature can still reach 0.97, 0.97, 0.94, and 0.72 after 400 iterations. Addition of a surface flux model improves retrieval quality. Allowing height-dependent eddy viscosity and diffusivity does not improve retrieval quality, whereas doubling the value of eddy diffusivity improves retrieval quality. Implementation of temporal and spatial smoothness penalty functions significantly improves retrieval quality in the presence of various sources of error.