In order to understand the spatial variability of soil respiration and its influencing factors, soil respiration (R_s) and related factors including soil temperature at 10 cm depth (T₁₀), soil water content from 0 to 10 cm depth (W_s), litter fall mass (L_w), litter fall moisture (L_m), soil total carbon (C), soil total nitrogen (N), and soil sulfur (S) were determined at 4, 2, and 1 m sampling scales, in a conifer-broadleaf mixed forest in Pangquangou nature reserve area of Shanxi province. The goals of this investigation were to monitor R_s heterogeneities at the stand scale and determine the correlations between R_s and affecting factors. The results from traditional statistics indicated that coefficients of variation (CV) for most of the measured factors ranged between 15%-59%; T₁₀ and C/N ratio were approximately 10%. The spatial variations of both T₁₀ and C/N ratio had low variability (CV ≤ 10%), and the others had medium variability (10% ≤ CV ≤ 100%). The simple linear correlations between R_s and L_w, L_m, C, N, and T₁₀ were all highly significant (P < 0.01); the correlation between R_s and W_s was significant (P < 0.05). However, no significant relationship between R_s and C/N ratio and S (P > 0.05) was observed. Stepwise multivariate regression demonstrated that the four factors of L_w, T₁₀, N, and C/N ratio together accounted for 26% of R_s heterogeneity, with the equation R_s = 11.972 + 0.033L_w- 0.267T₁₀ + 8.058N-0.390C/N (R² = 0.26, P = 0.000). Principal component analysis showed that the soil substrates of C and N, the environmental factors of T₁₀ and L_m, and the biotic factor of L_w could account for more than 70% of the spatial variation in R_s. The results from the geo-statistical analysis showed that the environmental factors of T₁₀, W_s, L_m, C, N, and C/N ratio had a significant spatial autocorrelation, and that structural factors played a leading role in their heterogeneity. S had a weak spatial autocorrelation, showing a random factor acted on its heterogeneity. The range of the semi-variogram function was about 17 m for R_s and the influencing factors. Fractal dimension was used to measure the complexity of natural phenomena, and the rank for the selected factors was in the following order: L_w (1.87) > S (1.84) > L_m (1.82) > N (1.77) > R_s (1.74) > C (1.73) > W_s (1.69) > T₁₀ (1.56) > C/N ratio (1.46). The spatial distribution model of R_s showed a similar pattern to that of W_s, L_m, L_w, C, N, and S, but not similar to that of T₁₀. The required sampling numbers of the R_s for 4, 2, and 1 m scales within ±5% and ±10% of its actual mean at the 95% confidence level were 74, 44, and 39, and 19, 11, and 10, respectively. This showed a decrease in the required sampling number coinciding with a decrease in the sampling scale; there was a similar trend in the estimation of accuracy. Our research results may have important applications in the study of CO₂ efflux in similar semiarid regions.