Using the Sloan Digital Sky Survey Data Release 5, we have investigated the intrinsic axis ratio distribution (ARD) for early-type galaxies. We have constructed a volume-limited sample of 3922 visually inspected early-type galaxies at 0.05 ≤ z ≤ 0.06, carefully considering sampling biases caused by the galaxy isophotal size and luminosity. We attempt to deproject the observed ARD into three-dimensional types (oblate, prolate, and triaxial), which are classified in terms of triaxiality. We confirm that no linear combination of randomly distributed axis ratios of the three types can reproduce the observed ARD. However, using Gaussian intrinsic distributions, we have found reasonable fits to the data using preferred mean axis ratios for the oblate, prolate, and triaxial (triaxials are given in two axis ratios) types of μo = 0.44, μp = 0.72, μt,β = 0.92, and μt,γ = 0.78, where the fractions of oblate, prolate, and triaxial types are O : P : T = 0.29 ± 0.09 : 0.26 ± 0.11 : 0.45 ± 0.13. We have also found that the luminous sample (-23.3 < Mr ≤ -21.2) tends to have more triaxial types than the less luminous (-21.2 < Mr < -19.3) sample does. Oblate types are relatively more abundant among the less luminous galaxies. Interestingly, the preferences of axis ratios for triaxial types in the two luminosity classes are remarkably similar. We have not found any significant influence of the local galaxy number density on the ARD. We show that the results can be seriously affected by the details of the data selection and type classification scheme. Caveats and implications for galaxy formation are discussed.