Volcanic earthquakes can provide significant insight into physical processes acting at volcanoes, such as magma accumulation and the mechanisms of deformation of the volcanic edifice. At the same time a statistical analyses of volcanic seismicity prior to an eruption (for example variations in the Gutenberg-Richter b-value – a measure of the proportion of large and small events) are a key component of the practical problem of forecasting eruptions. This thesis aims to tackle two key areas of research that are closely related to these important overall goals, by comparing seismic data obtained from currently-active volcanoes with direct field observation of faulting and fracturing from an exhumed extinct volcano. First I introduce a new approach that improves the accuracy and reliability of calculating spatial and temporal variations of the seismic b-value for frequency-magnitude distributions at active volcanoes, and apply it to several test cases. An extensive literature review highlights a large variability and lack of standardisation of methodology used to analyse frequency-magnitude distributions in the past. Motivated by this, I introduce and test a new workflow to standardise calculating completeness magnitudes of seismic catalogues. The review also highlights the fact that uncertainties in estimating the threshold magnitude of complete reporting have been ignored to date. Here I use synthetic catalogues to quantify this previously unidentified source of error, and provide a template to estimate the total error in b-value. In standard analysis it is also common to sample time windows subjectively, although this can introduce bias. Here I develop a new objective, iterative sampling method that calculates the b-value as a full probability density function which need not have a Gaussian error structure. Application of this method reveals ‘mode-switching’ behaviour for the first time in volcanic seismic catalogues. The results also show b-values often do have a value indistinguishable from that of tectonic seismicity (b=1 within error). Nevertheless there are also several robust examples of real high b-values, as high as 3.3. The second part of the study is based on a field campaign to investigate the fracture zones from an exhumed volcanic setting on the Isle of Rum, NW Scotland. Lithological and structural mapping is used to collect structural data that is then used to quantify and explain complex fracture patterns and the underlying intra-magma chamber processes that occurred there in the geological past. In particular I identify a singular collapse event within the youngest volcanic unit, the Central Intrusion. This is responsible for forming the observed igneous breccias and the lineaments on satellite images that I interpret as contemporaneous faults. Using appropriate scaling relations, I infer the b-value for the Rum lineaments data. This would have been relatively high, at a value of approximately 1.9. The final part of the study compares the fracture data on Rum to earthquake distributions at El Hierro volcano, Canary Islands. Here I show the level of fractal clustering is similar in both an extinct (60 Ma) and a currently active volcano. Both show similar high levels of clustering. However, in both cases there is a difference between the capacity and correlation dimensions (D₀≠D₂), implying the set of rupture sources or mapped fault traces form a multi-fractal set. Broadly, the scaling of fracture sets in an ancient volcano has similar properties to those observed in a modern volcano, except that the Rum data imply a greater absolute degree of spatial clustering of deformation than that for the recent unrest at El Hierro.