Spiking Neural Networks (SNNs), the third generation of neural networks, leverage biologically plausible spiking neurons and discrete spike-based encoding for efficient, robust, and energy-saving computation, especially suitable for neuromorphic hardware. However, current research lacks a fine-grained classification of spike encoding methods, hindering deeper understanding of their impact. This paper introduces a refined categorization based on spike sequence dimensionality, distinguishing univariate from multivariate encodings. We systematically evaluate various encoding strategies on three tasks: MNIST digit recognition, word similarity matching, and online handwriting recognition. Results reveal how encoding affects information representation, network optimization, and generalization. Code will be released at https://github.com/Liyzc/SNN_survey .