Electrocardiogram (ECG) signal has biomedical signs showing the working activity of the heart muscle by human body. Being able to take these biomedical signs from human body is important for diagnosis. Although there are a lot of Independent Component Analysis (ICA) algorithms that separate ECG signals from noises in the literature, separation qualities of them depend on initial conditions. The aim of this study is to find an ICA algorithm giving the stable separation quality for different initial conditions. Experimental results show that Kernel Independent Component Analysis algorithm (KICA) provides more accurate and stable results than the other algorithms.