We experimentally generate polarization spectroscopy (error) signals corresponding to the D2-line hyperfine transitions, Fg = 2 → Fe = 1, 2, 3 of 87Rb, and Fg = 3 → Fe = 2, 3, 4 of 85Rb, and show that the strongest error signals correspond to the closed hyperfine transitions (oscillator strength ∼ 0.7), Fg = 2 → Fe = 3 and Fg = 3 → Fe = 4, respectively. We make the generated error signal robust to fluctuations in external parameters by finding optimum values for the vapor cell temperature and pump-probe intensities at two different beam diameters. We further employ these optimized error signals to directly (without the need for frequency/phase modulation) lock the laser frequency —reducing the rms drift/linewidth from ∼ 10 MHz to < 500 kHz, when measured over a ∼ 60-minute duration. In addition, by comparing theoretically calculated and experimentally measured error signals for the pump intensities ranging from ∼ 0.1Isat0 − 10Isat0 (where, Isat0 ∼ 1.6 mW/cm 2 is the two-level saturation intensity for the Rb D2-line transitions), we discuss the applicability and limitations of existing numerical and analytical approaches based on a full multi-level rate equation model for polarization spectroscopy.