This study introduces a novel method for detecting and sorting spikes from multiunit recordings using wavelet transforms and superparamagnetic clustering (SPC). The method combines wavelet transforms, which localize spike features in time and frequency domains, with SPC, an unsupervised clustering technique that does not assume Gaussian distributions or low variance. The algorithm automatically sets amplitude thresholds for spike detection and uses wavelet coefficients to classify spikes. The entire process is unsupervised and efficient, making it suitable for large-scale neural recordings. The algorithm was tested on simulated data sets that closely resemble real recordings. It outperformed conventional methods in spike detection and clustering, particularly in cases with overlapping spikes and similar spike shapes. The method uses wavelet coefficients to extract features that are more effective for clustering than traditional methods like principal component analysis (PCA) or K-means. The wavelet transform allows for the localization of spike features, while SPC automatically identifies clusters without prior assumptions about data distribution. The algorithm's performance was evaluated using simulated data with varying noise levels and spike shapes. It successfully separated three spike classes even in challenging conditions, such as overlapping spikes and similar spike shapes. The method's unsupervised nature and efficiency make it particularly useful for analyzing large numbers of simultaneously recorded neurons. The study also compared the method with other clustering algorithms, including K-means, and found that SPC outperformed them in cases where spike shapes were similar. The use of wavelet coefficients, selected based on a Kolmogorov-Smirnov test for normality, provided better separation of clusters than PCA or fixed spike features. The method's ability to handle non-Gaussian distributions and overlapping spikes makes it a robust solution for spike sorting in real-world conditions. Overall, the proposed method offers an effective and efficient approach for unsupervised spike detection and sorting in multiunit recordings.This study introduces a novel method for detecting and sorting spikes from multiunit recordings using wavelet transforms and superparamagnetic clustering (SPC). The method combines wavelet transforms, which localize spike features in time and frequency domains, with SPC, an unsupervised clustering technique that does not assume Gaussian distributions or low variance. The algorithm automatically sets amplitude thresholds for spike detection and uses wavelet coefficients to classify spikes. The entire process is unsupervised and efficient, making it suitable for large-scale neural recordings. The algorithm was tested on simulated data sets that closely resemble real recordings. It outperformed conventional methods in spike detection and clustering, particularly in cases with overlapping spikes and similar spike shapes. The method uses wavelet coefficients to extract features that are more effective for clustering than traditional methods like principal component analysis (PCA) or K-means. The wavelet transform allows for the localization of spike features, while SPC automatically identifies clusters without prior assumptions about data distribution. The algorithm's performance was evaluated using simulated data with varying noise levels and spike shapes. It successfully separated three spike classes even in challenging conditions, such as overlapping spikes and similar spike shapes. The method's unsupervised nature and efficiency make it particularly useful for analyzing large numbers of simultaneously recorded neurons. The study also compared the method with other clustering algorithms, including K-means, and found that SPC outperformed them in cases where spike shapes were similar. The use of wavelet coefficients, selected based on a Kolmogorov-Smirnov test for normality, provided better separation of clusters than PCA or fixed spike features. The method's ability to handle non-Gaussian distributions and overlapping spikes makes it a robust solution for spike sorting in real-world conditions. Overall, the proposed method offers an effective and efficient approach for unsupervised spike detection and sorting in multiunit recordings.