The paper presents two non-linear methods, namely the fractal and the Konhonen net, used in the study of the urban system in Yangtze delta. Primacy index and the distribution curve of city scale in Yangtze delta are measured. Furthermore, the dimension of the scale structure is worked out to reveal the scale distribution in Yangtze delta. Then, spatial correlation dimension of the town-size distribution in Yangtze delta is also measured to analyses the spatial structure of urban system . In addition, based on the Kohonen net, using the data of the urban work force structure, all cities in Yangtze delta are classified into 5 groups in terms of the comprehensive urban function.
    According all above analyses it points out three character and deficiencies of the urban system in Yangtze delta, such as：（1）The scale distributing in Yangtze delta is mainly distributed with city-size distributing, partly possessing the characteristics of the first placed distribution；（2）The traffic networks in Yangtze delta does not coordinate with the urban system very much, and the capacity of spatial interaction is a little infirm；（3）Most cities are all-around cities, and the travel function is extremely outstanding in over half cities. But a few middle-small cities lack the special urban function.
    Some advice on developing the urban system of Yangtze delta is proposed based on the studying result：The integration of regional should be realized to form regional integration advantage, leading the whole economic development in delta. Such as：（1）Accelerate the step of the internationalization of Shanghai, develop the economy of big-cities, and the integration construction of the three core cities should be further promoted；（2）Perfect the traffic networks and communication networks to strengthen the relation among cities；（3）Realize the effective arrange of the urban function in Yangtze delta；（4）Develop the industry of travel and accelerate the intercity collaboration to build a big tourism circle. At last, the fractal theory was affirmed and the strongpoint of the Kohonen net was presented, namely its unsupervised learning. On the other hand, the improvement method is also put forward.