This dissertation presents a novel methodology for carrying out heart pathology diagnosis through similarity analysis of heart auscultation / sound (HS) signal using descriptive string complexity and inference networks technique. The proposed approach treats HS signal as a whole and segments the entire HS cycle for analysis, thereby avoids or solves the formidable problems existing in HS component segmentation and feature extraction in the traditional approach of HS analysis. The proposed methodology is based on the concept adapted from the general diagnosis procedure in clinical medicine. It includes (i) the concept of staged decision making process and (ii) the practice of comparing pathological symptoms of patients with well-known diagnosis samples. Technically, in this dissertation a new methodology is proposed to algorithmically describe HS signal and compare it against a HS signal with known pathology; subsequently inference nets are adapted to execute the staged decision-making process and generate the prognosis conclusion. Analyses of (dis)similarity between the information content of HS signals is then used to categorize the unknown HS signal into a particular cardiovascular disease (CVD) class. To algorithmically describe the HS signals, we investigate two parsing-based estimators, the Lempel-Ziv production complexity and Titchener’s T-information. Subsequently, a mathematical inference model is defined using our proposed conditional entropy estimator measure to analyse the (dis)similarity between the HS signals and classify the heart pathology. To identify the information content of HS signals for similarity analysis, feature vectors extracted from Mel-scale Frequency Cepstral Coefficients (MFCC), Musical Instrument Digital Interface (MIDI) and Pulse Code Modulated (PCM) sound data iv are investigated. It evaluates the three major types of feature of HS signals, namely the parametric feature – MFCC, morphological / acoustical feature – MIDI, and shape pattern – PCM, allowing one to define a suitable coding and encoding descriptor to analyse HS signals using our proposed methodology. Finally, with similarity analysis results, hierarchical inference nets are constructed to classify the heart pathology. A Generalized Inference Nets Model (GINM) is developed to ease the construction of hierarchical multistage inference nets for various applications. The generalization ability of GINM is validated by constructing inference nets to diagnose CVD using hemodynamic parameters (HDPs) derived from arterial pulse wave (APW). The results are further validated using real HS samples collected from Johns Hopkins hospital, USA and AWP signals acquired from two hospitals in China. The results suggest that the proposed approaches to CVD diagnosis using HS signal analysis and HDPs respectively could form a feasible technique in developing automate diagnosis system for e-home healthcare.