Current anomaly detection methods face two core challenges in tabular data: hypothesis misalignment and computational inefficiency. To address this, we propose the Sample Spatial Distribution and Diffusion Time Deviation Based Anomaly Detection Method (SSDADM). First, SSDADM employs a Sample Diffusion Model (SDM) to capture the spatial distribution difference of samples. It leverages a sample diffusion process (add sample to noise) and is trained using a sample prediction error loss, enabling efficient single-pass reconstruction to compute the reconstruction error as the spatial anomaly score. Then, the Sample Diffusion Time Network (SDTN) is introduced to assess diffusion time deviation. It is trained with time-consistent labels generated via the sample diffusion process, minimizing a time-consistent loss to output the diffusion time prediction as the temporal anomaly score. Finally, SSDADM uses the Distribution-Time Anomaly Score (DTAS) for anomaly detection. DTAS integrates the reconstruction error of SDM (reflecting spatial distribution difference) and the diffusion time prediction of SDTN (capturing diffusion time deviation). This joint scoring mechanism forms a multidimensional anomaly score, overcoming the accuracy limitation of single-metric methods. Experiments on 57 datasets demonstrate the effectiveness and efficiency of SSDADM compared to state-of-the-art methods.