The interrelationship between pattern and scale is still an important topic in landscape ecology。Scale means “the window of perception”, or the measurement of time, space and analysis (or observation). In recent years, the relationship between pattern and temporal or spatial scale has been extensively studied, but the relationship between pattern and analysis scale has been less studied. Usually landscape pattern is expressed by a series of pattern indices, which are counted under the control of the principle of thematic classification, so numbers of thematic classes may influence the value of indices. For example, if there is fewer classes (less than 5 classes) in one thematic map, many landscape patches are aggregated into one class, the landscape heterogeneity would be largely ignored, otherwise, the landscape boundary would be vague with too detailed classification. In this paper, analysis scale refers to the thematic resolution of spatial information or numbers of thematic classes in a map, and reflects the mapping detail of geo-information. The paper is expected to derive the proper number of thematic classes during landscape mapping and reveal how thematic resolution influences the landscape pattern.
 We design three typical districts as the research areas. One is Fuzhou administrative district representing the mixed landscapes with cities, forest and agricultural land under the similar land use policy and management in large region, next is Fuzhou urban area characterized by the human-made landscape at middle space scale, and the last one is Yongtai county representing the semi-natural landscape with a large proportion of forest & agricultural land at small space scale. We analyze the influence of NDVI（Normalized Difference Vegetation Index）classes on its spatial pattern. The pattern indices include NP (Number of Patches), LPI (Largest Patch Index), AREA-MN (Patch Area Mean), SHAPE-AM (Area-weighted Mean SHAPE Index), FRAC-AM (Area-weighted Mean patch Fractal Dimension), CONTAG (Contagion Index), IJI (Interspersion-Juxtaposition Index), and SHDI (Shannon′s Diversity Index). The analysis generated the following results.
Firstly, it is shown that there are critical analysis scale domains for pattern indices: if the number of classes is less than 4, indices change dramatically; if the number of classes is more than 12, indices keep stable, and the result is similar with that of the response of pattern indices to spatial scales. Moreover, the response sequence of pattern indices to analysis scale is identified: the initiatively sensitive scales (2—4 classes), sensitive scales (4—8 classes), proper analytical scales (8—12 classes), and insensitive scales (more than 12 classes).
 Secondly, by comparing the two group regions (Fuzhou administrative district / Fuzhou urban area, Fuzhou urban area / Yongtai county), except SHAPE-AM, the response trend of other pattern indices to the number of chasses keeps similar in despite of the changed values of pattern indices in different regions. Therefore, factors such as study area and human disturbance have little influence on the relationship of pattern-analysis scale.
 Lastly, the quantitative relationships between different pattern indices and thematic resolutions are listed as following: logarithmic growth for NP and SHDI with analysis scale, linear growth for LPI and CONTAG with analysis scale, power decrease for AREA-MN with analysis scale. The impact of thematic resolution on NDVI pattern is a result of the inherent hierarchy of the landscape, just like that of the spatial scale.
Overall, in a NDVI classification map, 8—12 classes are optimally recommended in the light of the landscape pattern. NDVI landscape pattern has steadily thematic scale-dependency without the influence of study area and strength of human activity. NP、SHDI and CONTAG should be selected as the typical pattern indices in that they steadily change with the analysis scales.